From 927553f32ad59f5914a1058ffea0259df6494f1a Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Wed, 30 Oct 2024 22:28:09 +0200
Subject: [PATCH 01/45] initial commit

---
 .../community/pipeline_flux_rf_inversion.py   | 933 ++++++++++++++++++
 1 file changed, 933 insertions(+)
 create mode 100644 examples/community/pipeline_flux_rf_inversion.py

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
new file mode 100644
index 000000000000..75c7813c7513
--- /dev/null
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -0,0 +1,933 @@
+# Copyright 2024 Black Forest Labs and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Union
+
+import numpy as np
+import torch
+from transformers import CLIPTextModel, CLIPTokenizer, T5EncoderModel, T5TokenizerFast
+
+from diffusers.image_processor import VaeImageProcessor
+from diffusers.loaders import FluxLoraLoaderMixin, FromSingleFileMixin
+from diffusers.models.autoencoders import AutoencoderKL
+from diffusers.models.transformers import FluxTransformer2DModel
+from diffusers.pipelines.flux.pipeline_output import FluxPipelineOutput
+from diffusers.pipelines.pipeline_utils import DiffusionPipeline
+from diffusers.schedulers import FlowMatchEulerDiscreteScheduler
+from diffusers.utils import (
+    USE_PEFT_BACKEND,
+    is_torch_xla_available,
+    logging,
+    replace_example_docstring,
+    scale_lora_layers,
+    unscale_lora_layers,
+)
+
+
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```py
+        >>> import torch
+        >>> from diffusers import FluxPipeline
+
+        >>> pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", torch_dtype=torch.bfloat16)
+        >>> pipe.to("cuda")
+        >>> prompt = "A cat holding a sign that says hello world"
+        >>> # Depending on the variant being used, the pipeline call will slightly vary.
+        >>> # Refer to the pipeline documentation for more details.
+        >>> image = pipe(prompt, num_inference_steps=4, guidance_scale=0.0).images[0]
+        >>> image.save("flux.png")
+        ```
+"""
+
+
+def calculate_shift(
+    image_seq_len,
+    base_seq_len: int = 256,
+    max_seq_len: int = 4096,
+    base_shift: float = 0.5,
+    max_shift: float = 1.16,
+):
+    m = (max_shift - base_shift) / (max_seq_len - base_seq_len)
+    b = base_shift - m * base_seq_len
+    mu = image_seq_len * m + b
+    return mu
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    sigmas: Optional[List[float]] = None,
+    **kwargs,
+):
+    r"""
+    Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+    custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+    Args:
+        scheduler (`SchedulerMixin`):
+            The scheduler to get timesteps from.
+        num_inference_steps (`int`):
+            The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+            must be `None`.
+        device (`str` or `torch.device`, *optional*):
+            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        timesteps (`List[int]`, *optional*):
+            Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+            `num_inference_steps` and `sigmas` must be `None`.
+        sigmas (`List[float]`, *optional*):
+            Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+            `num_inference_steps` and `timesteps` must be `None`.
+
+    Returns:
+        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+        second element is the number of inference steps.
+    """
+    if timesteps is not None and sigmas is not None:
+        raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+    if timesteps is not None:
+        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accepts_timesteps:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" timestep schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    elif sigmas is not None:
+        accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accept_sigmas:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" sigmas schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
+class RFInversionFluxPipeline(
+    DiffusionPipeline,
+    FluxLoraLoaderMixin,
+    FromSingleFileMixin,
+    TextualInversionLoaderMixin,
+):
+    r"""
+    The Flux pipeline for text-to-image generation.
+
+    Reference: https://blackforestlabs.ai/announcing-black-forest-labs/
+
+    Args:
+        transformer ([`FluxTransformer2DModel`]):
+            Conditional Transformer (MMDiT) architecture to denoise the encoded image latents.
+        scheduler ([`FlowMatchEulerDiscreteScheduler`]):
+            A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
+        vae ([`AutoencoderKL`]):
+            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+        text_encoder ([`CLIPTextModel`]):
+            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+        text_encoder_2 ([`T5EncoderModel`]):
+            [T5](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5EncoderModel), specifically
+            the [google/t5-v1_1-xxl](https://huggingface.co/google/t5-v1_1-xxl) variant.
+        tokenizer (`CLIPTokenizer`):
+            Tokenizer of class
+            [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer).
+        tokenizer_2 (`T5TokenizerFast`):
+            Second Tokenizer of class
+            [T5TokenizerFast](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5TokenizerFast).
+    """
+
+    model_cpu_offload_seq = "text_encoder->text_encoder_2->transformer->vae"
+    _optional_components = []
+    _callback_tensor_inputs = ["latents", "prompt_embeds"]
+
+    def __init__(
+        self,
+        scheduler: FlowMatchEulerDiscreteScheduler,
+        vae: AutoencoderKL,
+        text_encoder: CLIPTextModel,
+        tokenizer: CLIPTokenizer,
+        text_encoder_2: T5EncoderModel,
+        tokenizer_2: T5TokenizerFast,
+        transformer: FluxTransformer2DModel,
+    ):
+        super().__init__()
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            text_encoder_2=text_encoder_2,
+            tokenizer=tokenizer,
+            tokenizer_2=tokenizer_2,
+            transformer=transformer,
+            scheduler=scheduler,
+        )
+        self.vae_scale_factor = (
+            2 ** (len(self.vae.config.block_out_channels) - 1) if hasattr(self, "vae") and self.vae is not None else 8
+        )
+        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
+        self.tokenizer_max_length = (
+            self.tokenizer.model_max_length if hasattr(self, "tokenizer") and self.tokenizer is not None else 77
+        )
+        self.default_sample_size = 128
+
+    def _get_t5_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]] = None,
+        num_images_per_prompt: int = 1,
+        max_sequence_length: int = 512,
+        device: Optional[torch.device] = None,
+        dtype: Optional[torch.dtype] = None,
+    ):
+        device = device or self._execution_device
+        dtype = dtype or self.text_encoder.dtype
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        batch_size = len(prompt)
+
+        if isinstance(self, TextualInversionLoaderMixin):
+            prompt = self.maybe_convert_prompt(prompt, self.tokenizer_2)
+
+        text_inputs = self.tokenizer_2(
+            prompt,
+            padding="max_length",
+            max_length=max_sequence_length,
+            truncation=True,
+            return_length=False,
+            return_overflowing_tokens=False,
+            return_tensors="pt",
+        )
+        text_input_ids = text_inputs.input_ids
+        untruncated_ids = self.tokenizer_2(prompt, padding="longest", return_tensors="pt").input_ids
+
+        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
+            removed_text = self.tokenizer_2.batch_decode(untruncated_ids[:, self.tokenizer_max_length - 1 : -1])
+            logger.warning(
+                "The following part of your input was truncated because `max_sequence_length` is set to "
+                f" {max_sequence_length} tokens: {removed_text}"
+            )
+
+        prompt_embeds = self.text_encoder_2(text_input_ids.to(device), output_hidden_states=False)[0]
+
+        dtype = self.text_encoder_2.dtype
+        prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+
+        _, seq_len, _ = prompt_embeds.shape
+
+        # duplicate text embeddings and attention mask for each generation per prompt, using mps friendly method
+        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+        prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+        return prompt_embeds
+
+    def _get_clip_prompt_embeds(
+        self,
+        prompt: Union[str, List[str]],
+        num_images_per_prompt: int = 1,
+        device: Optional[torch.device] = None,
+    ):
+        device = device or self._execution_device
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+        batch_size = len(prompt)
+
+        if isinstance(self, TextualInversionLoaderMixin):
+            prompt = self.maybe_convert_prompt(prompt, self.tokenizer)
+
+        text_inputs = self.tokenizer(
+            prompt,
+            padding="max_length",
+            max_length=self.tokenizer_max_length,
+            truncation=True,
+            return_overflowing_tokens=False,
+            return_length=False,
+            return_tensors="pt",
+        )
+
+        text_input_ids = text_inputs.input_ids
+        untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
+            removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer_max_length - 1 : -1])
+            logger.warning(
+                "The following part of your input was truncated because CLIP can only handle sequences up to"
+                f" {self.tokenizer_max_length} tokens: {removed_text}"
+            )
+        prompt_embeds = self.text_encoder(text_input_ids.to(device), output_hidden_states=False)
+
+        # Use pooled output of CLIPTextModel
+        prompt_embeds = prompt_embeds.pooler_output
+        prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device)
+
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt)
+        prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, -1)
+
+        return prompt_embeds
+
+    def encode_prompt(
+        self,
+        prompt: Union[str, List[str]],
+        prompt_2: Union[str, List[str]],
+        device: Optional[torch.device] = None,
+        num_images_per_prompt: int = 1,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
+        max_sequence_length: int = 512,
+        lora_scale: Optional[float] = None,
+    ):
+        r"""
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                prompt to be encoded
+            prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+                used in all text-encoders
+            device: (`torch.device`):
+                torch device
+            num_images_per_prompt (`int`):
+                number of images that should be generated per prompt
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+                If not provided, pooled text embeddings will be generated from `prompt` input argument.
+            lora_scale (`float`, *optional*):
+                A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+        """
+        device = device or self._execution_device
+
+        # set lora scale so that monkey patched LoRA
+        # function of text encoder can correctly access it
+        if lora_scale is not None and isinstance(self, FluxLoraLoaderMixin):
+            self._lora_scale = lora_scale
+
+            # dynamically adjust the LoRA scale
+            if self.text_encoder is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder, lora_scale)
+            if self.text_encoder_2 is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder_2, lora_scale)
+
+        prompt = [prompt] if isinstance(prompt, str) else prompt
+
+        if prompt_embeds is None:
+            prompt_2 = prompt_2 or prompt
+            prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2
+
+            # We only use the pooled prompt output from the CLIPTextModel
+            pooled_prompt_embeds = self._get_clip_prompt_embeds(
+                prompt=prompt,
+                device=device,
+                num_images_per_prompt=num_images_per_prompt,
+            )
+            prompt_embeds = self._get_t5_prompt_embeds(
+                prompt=prompt_2,
+                num_images_per_prompt=num_images_per_prompt,
+                max_sequence_length=max_sequence_length,
+                device=device,
+            )
+
+        if self.text_encoder is not None:
+            if isinstance(self, FluxLoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder, lora_scale)
+
+        if self.text_encoder_2 is not None:
+            if isinstance(self, FluxLoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder_2, lora_scale)
+
+        dtype = self.text_encoder.dtype if self.text_encoder is not None else self.transformer.dtype
+        text_ids = torch.zeros(prompt_embeds.shape[1], 3).to(device=device, dtype=dtype)
+
+        return prompt_embeds, pooled_prompt_embeds, text_ids
+
+    @torch.no_grad()
+    # Modified from diffusers.pipelines.ledits_pp.pipeline_leditspp_stable_diffusion.LEditsPPPipelineStableDiffusion.encode_image
+    def encode_image(self, image, dtype=None, height=None, width=None, resize_mode="default", crops_coords=None):
+        image = self.image_processor.preprocess(
+            image=image, height=height, width=width, resize_mode=resize_mode, crops_coords=crops_coords
+        )
+        resized = self.image_processor.postprocess(image=image, output_type="pil")
+
+        if max(image.shape[-2:]) > self.vae.config["sample_size"] * 1.5:
+            logger.warning(
+                "Your input images far exceed the default resolution of the underlying diffusion model. "
+                "The output images may contain severe artifacts! "
+                "Consider down-sampling the input using the `height` and `width` parameters"
+            )
+        image = image.to(dtype)
+
+        x0 = self.vae.encode(image.to(self.device)).latent_dist.sample()
+        x0 = (x0 - self.vae.config.shift_factor) * self.vae.config.scaling_factor
+        x0 = x0.to(dtype)
+        return x0, resized
+    def check_inputs(
+        self,
+        prompt,
+        prompt_2,
+        height,
+        width,
+        prompt_embeds=None,
+        pooled_prompt_embeds=None,
+        callback_on_step_end_tensor_inputs=None,
+        max_sequence_length=None,
+    ):
+        if height % self.vae_scale_factor != 0 or width % self.vae_scale_factor != 0:
+            raise ValueError(
+                f"`height` and `width` have to be divisible by {self.vae_scale_factor} but are {height} and {width}."
+            )
+
+        if callback_on_step_end_tensor_inputs is not None and not all(
+            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+        ):
+            raise ValueError(
+                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+            )
+
+        if prompt is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt_2 is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+        elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+            raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+        if prompt_embeds is not None and pooled_prompt_embeds is None:
+            raise ValueError(
+                "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`."
+            )
+
+        if max_sequence_length is not None and max_sequence_length > 512:
+            raise ValueError(f"`max_sequence_length` cannot be greater than 512 but is {max_sequence_length}")
+
+    @staticmethod
+    def _prepare_latent_image_ids(batch_size, height, width, device, dtype):
+        latent_image_ids = torch.zeros(height, width, 3)
+        latent_image_ids[..., 1] = latent_image_ids[..., 1] + torch.arange(height)[:, None]
+        latent_image_ids[..., 2] = latent_image_ids[..., 2] + torch.arange(width)[None, :]
+
+        latent_image_id_height, latent_image_id_width, latent_image_id_channels = latent_image_ids.shape
+
+        latent_image_ids = latent_image_ids.reshape(
+            latent_image_id_height * latent_image_id_width, latent_image_id_channels
+        )
+
+        return latent_image_ids.to(device=device, dtype=dtype)
+
+    @staticmethod
+    def _pack_latents(latents, batch_size, num_channels_latents, height, width):
+        latents = latents.view(batch_size, num_channels_latents, height // 2, 2, width // 2, 2)
+        latents = latents.permute(0, 2, 4, 1, 3, 5)
+        latents = latents.reshape(batch_size, (height // 2) * (width // 2), num_channels_latents * 4)
+
+        return latents
+
+    @staticmethod
+    def _unpack_latents(latents, height, width, vae_scale_factor):
+        batch_size, num_patches, channels = latents.shape
+
+        height = height // vae_scale_factor
+        width = width // vae_scale_factor
+
+        latents = latents.view(batch_size, height // 2, width // 2, channels // 4, 2, 2)
+        latents = latents.permute(0, 3, 1, 4, 2, 5)
+
+        latents = latents.reshape(batch_size, channels // (2 * 2), height, width)
+
+        return latents
+
+    def enable_vae_slicing(self):
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.vae.enable_slicing()
+
+    def disable_vae_slicing(self):
+        r"""
+        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_slicing()
+
+    def enable_vae_tiling(self):
+        r"""
+        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+        processing larger images.
+        """
+        self.vae.enable_tiling()
+
+    def disable_vae_tiling(self):
+        r"""
+        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_tiling()
+
+    def prepare_latents(
+        self,
+        batch_size,
+        num_channels_latents,
+        height,
+        width,
+        dtype,
+        device,
+        generator,
+        latents=None,
+    ):
+        height = int(height) // self.vae_scale_factor
+        width = int(width) // self.vae_scale_factor
+
+        shape = (batch_size, num_channels_latents, height, width)
+
+        if latents is not None:
+            latent_image_ids = self._prepare_latent_image_ids(batch_size, height, width, device, dtype)
+            return latents.to(device=device, dtype=dtype), latent_image_ids
+
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+        latents = self._pack_latents(latents, batch_size, num_channels_latents, height, width)
+
+        latent_image_ids = self._prepare_latent_image_ids(batch_size, height // 2, width // 2, device, dtype)
+
+        return latents, latent_image_ids
+
+    def prepare_inverted_latents(
+        self,
+        batch_size,
+        num_channels_latents,
+        height,
+        width,
+        dtype,
+        device,
+        generator,
+    ):
+
+        latent_image_ids = self._prepare_latent_image_ids(batch_size, height, width, device, dtype)
+
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        latents = self._pack_latents(self.inverted_latents, batch_size, num_channels_latents, height, width)
+        return latents, latent_image_ids
+
+    @property
+    def guidance_scale(self):
+        return self._guidance_scale
+
+    @property
+    def joint_attention_kwargs(self):
+        return self._joint_attention_kwargs
+
+    @property
+    def num_timesteps(self):
+        return self._num_timesteps
+
+    @property
+    def interrupt(self):
+        return self._interrupt
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        prompt_2: Optional[Union[str, List[str]]] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_inference_steps: int = 28,
+        eta_base: float = 1.0,  # base eta value
+        eta_trend: str = 'linear_decrease',  # constant, linear_increase, linear_decrease
+        start_step: int = 0,  # 0-based indexing, closed interval
+        end_step: int = 30,  # 0-based indexing, open interval
+        timesteps: List[int] = None,
+        use_shift_t_sampling: bool =True,
+        guidance_scale: float = 3.5,
+        num_images_per_prompt: Optional[int] = 1,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
+        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        max_sequence_length: int = 512,
+    ):
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+                will be used instead
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The height in pixels of the generated image. This is set to 1024 by default for the best results.
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The width in pixels of the generated image. This is set to 1024 by default for the best results.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            timesteps (`List[int]`, *optional*):
+                Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+                in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+                passed will be used. Must be in descending order.
+            guidance_scale (`float`, *optional*, defaults to 7.0):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
+                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
+                usually at the expense of lower image quality.
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+                to make generation deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will ge generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+                If not provided, pooled text embeddings will be generated from `prompt` input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.flux.FluxPipelineOutput`] instead of a plain tuple.
+            joint_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            callback_on_step_end (`Callable`, *optional*):
+                A function that calls at the end of each denoising steps during the inference. The function is called
+                with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+                callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+                `callback_on_step_end_tensor_inputs`.
+            callback_on_step_end_tensor_inputs (`List`, *optional*):
+                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+                `._callback_tensor_inputs` attribute of your pipeline class.
+            max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.
+
+        Examples:
+
+        Returns:
+            [`~pipelines.flux.FluxPipelineOutput`] or `tuple`: [`~pipelines.flux.FluxPipelineOutput`] if `return_dict`
+            is True, otherwise a `tuple`. When returning a tuple, the first element is a list with the generated
+            images.
+        """
+
+        height = height or self.default_sample_size * self.vae_scale_factor
+        width = width or self.default_sample_size * self.vae_scale_factor
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            prompt_2,
+            height,
+            width,
+            prompt_embeds=prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
+            max_sequence_length=max_sequence_length,
+        )
+
+        self._guidance_scale = guidance_scale
+        self._joint_attention_kwargs = joint_attention_kwargs
+        self._interrupt = False
+
+        # 2. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        device = self._execution_device
+
+        lora_scale = (
+            self.joint_attention_kwargs.get("scale", None) if self.joint_attention_kwargs is not None else None
+        )
+        (
+            prompt_embeds,
+            pooled_prompt_embeds,
+            text_ids,
+        ) = self.encode_prompt(
+            prompt=prompt,
+            prompt_2=prompt_2,
+            prompt_embeds=prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            device=device,
+            num_images_per_prompt=num_images_per_prompt,
+            max_sequence_length=max_sequence_length,
+            lora_scale=lora_scale,
+        )
+
+        # 4. Prepare latent variables
+        num_channels_latents = self.transformer.config.in_channels // 4
+        # latents, latent_image_ids = self.prepare_latents(
+        #     batch_size * num_images_per_prompt,
+        #     num_channels_latents,
+        #     height,
+        #     width,
+        #     prompt_embeds.dtype,
+        #     device,
+        #     generator,
+        #     latents,
+        # )
+        packed_inv_latents, latent_image_ids = self.prepare_inverted_latents(
+            self.inverted_latents.shape[0],
+            self.inverted_latents.shape[1],
+            self.inverted_latents.shape[2],
+            self.inverted_latents.shape[3],
+            prompt_embeds.dtype,
+            device,
+            generator,
+        )
+        # 5. Prepare timesteps
+        sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
+        image_seq_len = latents.shape[1]
+        mu = calculate_shift(
+            image_seq_len,
+            self.scheduler.config.base_image_seq_len,
+            self.scheduler.config.max_image_seq_len,
+            self.scheduler.config.base_shift,
+            self.scheduler.config.max_shift,
+        )
+        timesteps, num_inference_steps = retrieve_timesteps(
+            self.scheduler,
+            num_inference_steps,
+            device,
+            timesteps,
+            sigmas,
+            mu=mu,
+        )
+        num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+        self._num_timesteps = len(timesteps)
+
+        # handle guidance
+        if self.transformer.config.guidance_embeds:
+            guidance = torch.full([1], guidance_scale, device=device, dtype=torch.float32)
+            guidance = guidance.expand(latents.shape[0])
+        else:
+            guidance = None
+
+        # 6. Denoising loop
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                if self.interrupt:
+                    continue
+
+                # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+                timestep = t.expand(latents.shape[0]).to(latents.dtype)
+
+                noise_pred = self.transformer(
+                    hidden_states=latents,
+                    timestep=timestep / 1000,
+                    guidance=guidance,
+                    pooled_projections=pooled_prompt_embeds,
+                    encoder_hidden_states=prompt_embeds,
+                    txt_ids=text_ids,
+                    img_ids=latent_image_ids,
+                    joint_attention_kwargs=self.joint_attention_kwargs,
+                    return_dict=False,
+                )[0]
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents_dtype = latents.dtype
+                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+
+                if latents.dtype != latents_dtype:
+                    if torch.backends.mps.is_available():
+                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                        latents = latents.to(latents_dtype)
+
+                if callback_on_step_end is not None:
+                    callback_kwargs = {}
+                    for k in callback_on_step_end_tensor_inputs:
+                        callback_kwargs[k] = locals()[k]
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                    latents = callback_outputs.pop("latents", latents)
+                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
+        if output_type == "latent":
+            image = latents
+
+        else:
+            latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
+            latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor
+            image = self.vae.decode(latents, return_dict=False)[0]
+            image = self.image_processor.postprocess(image, output_type=output_type)
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (image,)
+
+        return FluxPipelineOutput(images=image)
+
+    @torch.no_grad()
+    def invert(
+            self,
+            image: PipelineImageInput,
+            source_prompt: str = "",
+            source_guidance_scale = 0.0,
+            num_inversion_steps: int = 28,
+            gamma: float = 0.5,
+            use_shift_t_sampling: bool = False,
+            generator: Optional[torch.Generator] = None,
+            dtype: Optional[torch.dtype] = None,
+    ):
+        dtype = dtype or self.text_encoder.dtype
+        height = height or self.default_sample_size * self.vae_scale_factor
+        width = width or self.default_sample_size * self.vae_scale_factor
+        # 1. prepare image
+        img_latents, _ = self.encode_image(image, dtype=dtype)
+
+        timesteps = get_schedule(
+            num_steps=num_steps,
+            image_seq_len=(1024 // 16) * (1024 // 16),  # vae_scale_factor = 16
+            shift=use_shift_t_sampling,  # Set True for Flux-dev, False for Flux-schnell
+        )[::-1]  # flipped for inversion
+
+        prompt_embeds, pooled_prompt_embeds, text_ids = pipeline.encode_prompt(
+            prompt=source_prompt
+        )
+        latent_image_ids = self._prepare_latent_image_ids(
+            img_latents.shape[0],
+            img_latents.shape[2],
+            img_latents.shape[3],
+            img_latents.device,
+            dtype,
+        )
+        packed_latents = self._pack_latents(
+            latents,
+            batch_size=latents.shape[0],
+            num_channels_latents=latents.shape[1],
+            height=latents.shape[2],
+            width=latents.shape[3],
+        )
+
+        target_noise = torch.randn(packed_latents.shape, device=packed_latents.device, dtype=torch.float32)
+
+        guidance_vec = torch.full((packed_latents.shape[0],), source_guidance_scale, device=packed_latents.device,
+                                  dtype=packed_latents.dtype)
+
+        # Image inversion with interpolated velocity field.  t goes from 0.0 to 1.0
+        with self.progress_bar(total=len(timesteps) - 1) as progress_bar:
+            for t_curr, t_prev in zip(timesteps[:-1], timesteps[1:]):
+                t_vec = torch.full((packed_latents.shape[0],), t_curr, dtype=packed_latents.dtype,
+                                   device=packed_latents.device)
+
+                # Null text velocity
+                flux_velocity = pipeline.transformer(
+                    hidden_states=packed_latents,
+                    timestep=t_vec,
+                    guidance=guidance_vec,
+                    pooled_projections=pooled_prompt_embeds,
+                    encoder_hidden_states=prompt_embeds,
+                    txt_ids=text_ids,
+                    img_ids=latent_image_ids,
+                    joint_attention_kwargs=None,
+                    return_dict=pipeline,
+                )[0]
+
+                # Prevents precision issues
+                packed_latents = packed_latents.to(torch.float32)
+                flux_velocity = flux_velocity.to(torch.float32)
+
+                # Target noise velocity
+                target_noise_velocity = (target_noise - packed_latents) / (1.0 - t_curr)
+
+                # interpolated velocity
+                interpolated_velocity = gamma * target_noise_velocity + (1 - gamma) * flux_velocity
+
+                # one step Euler
+                packed_latents = packed_latents + (t_prev - t_curr) * interpolated_velocity
+
+                packed_latents = packed_latents.to(dtype)
+                progress_bar.update()
+
+        print("Mean Absolute Error", torch.mean(torch.abs(packed_latents - target_noise)))
+
+        latents = self._unpack_latents(
+            packed_latents,
+            height=height,
+            width=width,
+            vae_scale_factor=self.vae_scale_factor,
+        )
+        latents = latents.to(dtype)
+        self.inverted_latents = latents
+        self.image_latents = img_latents
+        return latents

From 9382687a859403523374aac845114a4e4b7a904a Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Thu, 31 Oct 2024 12:03:52 +0200
Subject: [PATCH 02/45] update denoising loop

---
 .../community/pipeline_flux_rf_inversion.py   | 74 ++++++++++++++++---
 1 file changed, 65 insertions(+), 9 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 75c7813c7513..78c18cba589f 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -746,6 +746,17 @@ def __call__(
             device,
             generator,
         )
+
+        packed_img_latents = self._pack_latents(
+            self.image_latents,
+            batch_size=self.image_latents.shape[0],
+            num_channels_latents=self.image_latents.shape[1],
+            height=self.image_latents.shape[2],
+            width=self.image_latents.shape[3],
+        )
+        target_img = packed_img_latents.clone().to(torch.float32)
+
+
         # 5. Prepare timesteps
         sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
         image_seq_len = latents.shape[1]
@@ -770,20 +781,22 @@ def __call__(
         # handle guidance
         if self.transformer.config.guidance_embeds:
             guidance = torch.full([1], guidance_scale, device=device, dtype=torch.float32)
-            guidance = guidance.expand(latents.shape[0])
+            guidance = guidance.expand(packed_inv_latents.shape[0])
         else:
             guidance = None
 
+        eta_values = self.generate_eta_values(timesteps, start_step, end_step, eta_base, eta_trend)
+
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:
-            for i, t in enumerate(timesteps):
-                if self.interrupt:
-                    continue
+            for idx, (t_curr, t_prev) in enumerate(zip(timesteps[:-1], timesteps[1:])):
+                timestep = torch.full((packed_inv_latents.shape[0],), t_curr, dtype=packed_inv_latents.dtype,
+                                   device=packed_inv_latents.device)
 
                 # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
-                timestep = t.expand(latents.shape[0]).to(latents.dtype)
+                #timestep = t.expand(latents.shape[0]).to(latents.dtype)
 
-                noise_pred = self.transformer(
+                velocity = self.transformer(
                     hidden_states=latents,
                     timestep=timestep / 1000,
                     guidance=guidance,
@@ -796,8 +809,22 @@ def __call__(
                 )[0]
 
                 # compute the previous noisy sample x_t -> x_t-1
-                latents_dtype = latents.dtype
-                latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+                #latents_dtype = latents.dtype
+                # Prevents precision issues
+                packed_inv_latents = packed_inv_latents.to(torch.float32)
+                velocity = velocity.to(torch.float32)
+
+                # Target image velocity
+                target_img_velocity = -(target_img - packed_inv_latents) / t_curr
+
+                # interpolated velocity
+                eta = eta_values[idx]
+                interpolated_velocity = eta * target_img_velocity + (1 - eta) * velocity
+                latents = packed_inv_latents + (t_prev - t_curr) * interpolated_velocity
+                print(
+                    f"X_{t_prev:.3f} = X_{t_curr:.3f} + {t_prev - t_curr:.3f} * ({eta:.3f} * target_img_velocity + {1 - eta:.3f} * flux_velocity)")
+
+                #latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
 
                 if latents.dtype != latents_dtype:
                     if torch.backends.mps.is_available():
@@ -837,6 +864,35 @@ def __call__(
 
         return FluxPipelineOutput(images=image)
 
+    def generate_eta_values(
+            self,
+            timesteps,
+            start_step,
+            end_step,
+            eta,
+            eta_trend,
+    ):
+        assert start_step < end_step and start_step >= 0 and end_step <= len(
+            timesteps), "Invalid start_step and end_step"
+        # timesteps are monotonically decreasing, from 1.0 to 0.0
+        eta_values = [0.0] * (len(timesteps) - 1)
+
+        if eta_trend == 'constant':
+            for i in range(start_step, end_step):
+                eta_values[i] = eta
+        elif eta_trend == 'linear_increase':
+            total_time = timesteps[start_step] - timesteps[end_step - 1]
+            for i in range(start_step, end_step):
+                eta_values[i] = eta * (timesteps[start_step] - timesteps[i]) / total_time
+        elif eta_trend == 'linear_decrease':
+            total_time = timesteps[start_step] - timesteps[end_step - 1]
+            for i in range(start_step, end_step):
+                eta_values[i] = eta * (timesteps[i] - timesteps[end_step - 1]) / total_time
+        else:
+            raise NotImplementedError(f"Unsupported eta_trend: {eta_trend}")
+
+        return eta_values
+
     @torch.no_grad()
     def invert(
             self,
@@ -913,7 +969,7 @@ def invert(
                 # interpolated velocity
                 interpolated_velocity = gamma * target_noise_velocity + (1 - gamma) * flux_velocity
 
-                # one step Euler
+                # one-step Euler
                 packed_latents = packed_latents + (t_prev - t_curr) * interpolated_velocity
 
                 packed_latents = packed_latents.to(dtype)

From ac497896cf08c1da7a36eeaff9a28b8866944527 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Thu, 31 Oct 2024 12:58:45 +0200
Subject: [PATCH 03/45] fix scheduling

---
 .../community/pipeline_flux_rf_inversion.py   | 62 +++++++++++++------
 1 file changed, 43 insertions(+), 19 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 78c18cba589f..da204df38b6c 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -759,7 +759,7 @@ def __call__(
 
         # 5. Prepare timesteps
         sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
-        image_seq_len = latents.shape[1]
+        image_seq_len = packed_inv_latents.shape[1]
         mu = calculate_shift(
             image_seq_len,
             self.scheduler.config.base_image_seq_len,
@@ -790,11 +790,10 @@ def __call__(
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:
             for idx, (t_curr, t_prev) in enumerate(zip(timesteps[:-1], timesteps[1:])):
-                timestep = torch.full((packed_inv_latents.shape[0],), t_curr, dtype=packed_inv_latents.dtype,
-                                   device=packed_inv_latents.device)
 
                 # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
                 #timestep = t.expand(latents.shape[0]).to(latents.dtype)
+                timestep = t_curr.expand(packed_inv_latents.shape[0]).to(packed_inv_latents.dtype)
 
                 velocity = self.transformer(
                     hidden_states=latents,
@@ -908,14 +907,33 @@ def invert(
         dtype = dtype or self.text_encoder.dtype
         height = height or self.default_sample_size * self.vae_scale_factor
         width = width or self.default_sample_size * self.vae_scale_factor
+        device = self._execution_device
+
         # 1. prepare image
         img_latents, _ = self.encode_image(image, dtype=dtype)
 
-        timesteps = get_schedule(
-            num_steps=num_steps,
-            image_seq_len=(1024 // 16) * (1024 // 16),  # vae_scale_factor = 16
-            shift=use_shift_t_sampling,  # Set True for Flux-dev, False for Flux-schnell
-        )[::-1]  # flipped for inversion
+        # 2. Prepare timesteps
+        sigmas = np.linspace(1.0, 1 / num_inversion_steps, num_inversion_steps)
+        image_seq_len = img_latents.shape[1]
+        if use_shift_t_sampling:
+            mu = calculate_shift(
+                image_seq_len,
+                self.scheduler.config.base_image_seq_len,
+                self.scheduler.config.max_image_seq_len,
+                self.scheduler.config.base_shift,
+                self.scheduler.config.max_shift,
+            )
+        else:
+            mu = None
+        timesteps, num_inversion_steps = retrieve_timesteps(
+            self.scheduler,
+            num_inversion_steps,
+            device,
+            sigmas=sigmas,
+            mu=mu,
+        )
+
+        timesteps = timesteps[::-1] # flip for inversion
 
         prompt_embeds, pooled_prompt_embeds, text_ids = pipeline.encode_prompt(
             prompt=source_prompt
@@ -924,7 +942,7 @@ def invert(
             img_latents.shape[0],
             img_latents.shape[2],
             img_latents.shape[3],
-            img_latents.device,
+            device,
             dtype,
         )
         packed_latents = self._pack_latents(
@@ -935,22 +953,28 @@ def invert(
             width=latents.shape[3],
         )
 
-        target_noise = torch.randn(packed_latents.shape, device=packed_latents.device, dtype=torch.float32)
+        target_noise = torch.randn(packed_latents.shape, device=device, dtype=torch.float32)
 
-        guidance_vec = torch.full((packed_latents.shape[0],), source_guidance_scale, device=packed_latents.device,
-                                  dtype=packed_latents.dtype)
+        # handle guidance
+        if self.transformer.config.guidance_embeds:
+            guidance = torch.full([1], source_guidance_scale, device=device, dtype=torch.float32)
+            guidance = guidance.expand(packed_latents.shape[0])
+        else:
+            guidance = None
 
         # Image inversion with interpolated velocity field.  t goes from 0.0 to 1.0
         with self.progress_bar(total=len(timesteps) - 1) as progress_bar:
             for t_curr, t_prev in zip(timesteps[:-1], timesteps[1:]):
-                t_vec = torch.full((packed_latents.shape[0],), t_curr, dtype=packed_latents.dtype,
-                                   device=packed_latents.device)
+
+                # t_vec = torch.full((packed_latents.shape[0],), t_curr, dtype=packed_latents.dtype,
+                #                    device=device)
+                timestep = t_curr.expand(packed_latents.shape[0]).to(packed_latents.dtype)
 
                 # Null text velocity
-                flux_velocity = pipeline.transformer(
+                velocity = pipeline.transformer(
                     hidden_states=packed_latents,
-                    timestep=t_vec,
-                    guidance=guidance_vec,
+                    timestep=timestep,
+                    guidance=guidance,
                     pooled_projections=pooled_prompt_embeds,
                     encoder_hidden_states=prompt_embeds,
                     txt_ids=text_ids,
@@ -961,13 +985,13 @@ def invert(
 
                 # Prevents precision issues
                 packed_latents = packed_latents.to(torch.float32)
-                flux_velocity = flux_velocity.to(torch.float32)
+                velocity = velocity.to(torch.float32)
 
                 # Target noise velocity
                 target_noise_velocity = (target_noise - packed_latents) / (1.0 - t_curr)
 
                 # interpolated velocity
-                interpolated_velocity = gamma * target_noise_velocity + (1 - gamma) * flux_velocity
+                interpolated_velocity = gamma * target_noise_velocity + (1 - gamma) * velocity
 
                 # one-step Euler
                 packed_latents = packed_latents + (t_prev - t_curr) * interpolated_velocity

From bd9b7f1913ae4ee7b9b391e97ecb6cb3475c9a6b Mon Sep 17 00:00:00 2001
From: Linoy <linoy@huggingface.co>
Date: Thu, 31 Oct 2024 12:25:27 +0000
Subject: [PATCH 04/45] style

---
 .../community/pipeline_flux_rf_inversion.py   | 65 +++++++++----------
 1 file changed, 31 insertions(+), 34 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index da204df38b6c..c3bf7a1def16 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -394,6 +394,7 @@ def encode_image(self, image, dtype=None, height=None, width=None, resize_mode="
         x0 = (x0 - self.vae.config.shift_factor) * self.vae.config.scaling_factor
         x0 = x0.to(dtype)
         return x0, resized
+
     def check_inputs(
         self,
         prompt,
@@ -552,7 +553,6 @@ def prepare_inverted_latents(
         device,
         generator,
     ):
-
         latent_image_ids = self._prepare_latent_image_ids(batch_size, height, width, device, dtype)
 
         if isinstance(generator, list) and len(generator) != batch_size:
@@ -590,11 +590,11 @@ def __call__(
         width: Optional[int] = None,
         num_inference_steps: int = 28,
         eta_base: float = 1.0,  # base eta value
-        eta_trend: str = 'linear_decrease',  # constant, linear_increase, linear_decrease
+        eta_trend: str = "linear_decrease",  # constant, linear_increase, linear_decrease
         start_step: int = 0,  # 0-based indexing, closed interval
         end_step: int = 30,  # 0-based indexing, open interval
         timesteps: List[int] = None,
-        use_shift_t_sampling: bool =True,
+        use_shift_t_sampling: bool = True,
         guidance_scale: float = 3.5,
         num_images_per_prompt: Optional[int] = 1,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
@@ -756,7 +756,6 @@ def __call__(
         )
         target_img = packed_img_latents.clone().to(torch.float32)
 
-
         # 5. Prepare timesteps
         sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
         image_seq_len = packed_inv_latents.shape[1]
@@ -790,9 +789,8 @@ def __call__(
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:
             for idx, (t_curr, t_prev) in enumerate(zip(timesteps[:-1], timesteps[1:])):
-
                 # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
-                #timestep = t.expand(latents.shape[0]).to(latents.dtype)
+                # timestep = t.expand(latents.shape[0]).to(latents.dtype)
                 timestep = t_curr.expand(packed_inv_latents.shape[0]).to(packed_inv_latents.dtype)
 
                 velocity = self.transformer(
@@ -808,7 +806,7 @@ def __call__(
                 )[0]
 
                 # compute the previous noisy sample x_t -> x_t-1
-                #latents_dtype = latents.dtype
+                # latents_dtype = latents.dtype
                 # Prevents precision issues
                 packed_inv_latents = packed_inv_latents.to(torch.float32)
                 velocity = velocity.to(torch.float32)
@@ -821,9 +819,10 @@ def __call__(
                 interpolated_velocity = eta * target_img_velocity + (1 - eta) * velocity
                 latents = packed_inv_latents + (t_prev - t_curr) * interpolated_velocity
                 print(
-                    f"X_{t_prev:.3f} = X_{t_curr:.3f} + {t_prev - t_curr:.3f} * ({eta:.3f} * target_img_velocity + {1 - eta:.3f} * flux_velocity)")
+                    f"X_{t_prev:.3f} = X_{t_curr:.3f} + {t_prev - t_curr:.3f} * ({eta:.3f} * target_img_velocity + {1 - eta:.3f} * flux_velocity)"
+                )
 
-                #latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+                # latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
 
                 if latents.dtype != latents_dtype:
                     if torch.backends.mps.is_available():
@@ -864,26 +863,27 @@ def __call__(
         return FluxPipelineOutput(images=image)
 
     def generate_eta_values(
-            self,
-            timesteps,
-            start_step,
-            end_step,
-            eta,
-            eta_trend,
+        self,
+        timesteps,
+        start_step,
+        end_step,
+        eta,
+        eta_trend,
     ):
-        assert start_step < end_step and start_step >= 0 and end_step <= len(
-            timesteps), "Invalid start_step and end_step"
+        assert (
+            start_step < end_step and start_step >= 0 and end_step <= len(timesteps)
+        ), "Invalid start_step and end_step"
         # timesteps are monotonically decreasing, from 1.0 to 0.0
         eta_values = [0.0] * (len(timesteps) - 1)
 
-        if eta_trend == 'constant':
+        if eta_trend == "constant":
             for i in range(start_step, end_step):
                 eta_values[i] = eta
-        elif eta_trend == 'linear_increase':
+        elif eta_trend == "linear_increase":
             total_time = timesteps[start_step] - timesteps[end_step - 1]
             for i in range(start_step, end_step):
                 eta_values[i] = eta * (timesteps[start_step] - timesteps[i]) / total_time
-        elif eta_trend == 'linear_decrease':
+        elif eta_trend == "linear_decrease":
             total_time = timesteps[start_step] - timesteps[end_step - 1]
             for i in range(start_step, end_step):
                 eta_values[i] = eta * (timesteps[i] - timesteps[end_step - 1]) / total_time
@@ -894,15 +894,15 @@ def generate_eta_values(
 
     @torch.no_grad()
     def invert(
-            self,
-            image: PipelineImageInput,
-            source_prompt: str = "",
-            source_guidance_scale = 0.0,
-            num_inversion_steps: int = 28,
-            gamma: float = 0.5,
-            use_shift_t_sampling: bool = False,
-            generator: Optional[torch.Generator] = None,
-            dtype: Optional[torch.dtype] = None,
+        self,
+        image: PipelineImageInput,
+        source_prompt: str = "",
+        source_guidance_scale=0.0,
+        num_inversion_steps: int = 28,
+        gamma: float = 0.5,
+        use_shift_t_sampling: bool = False,
+        generator: Optional[torch.Generator] = None,
+        dtype: Optional[torch.dtype] = None,
     ):
         dtype = dtype or self.text_encoder.dtype
         height = height or self.default_sample_size * self.vae_scale_factor
@@ -933,11 +933,9 @@ def invert(
             mu=mu,
         )
 
-        timesteps = timesteps[::-1] # flip for inversion
+        timesteps = timesteps[::-1]  # flip for inversion
 
-        prompt_embeds, pooled_prompt_embeds, text_ids = pipeline.encode_prompt(
-            prompt=source_prompt
-        )
+        prompt_embeds, pooled_prompt_embeds, text_ids = pipeline.encode_prompt(prompt=source_prompt)
         latent_image_ids = self._prepare_latent_image_ids(
             img_latents.shape[0],
             img_latents.shape[2],
@@ -965,7 +963,6 @@ def invert(
         # Image inversion with interpolated velocity field.  t goes from 0.0 to 1.0
         with self.progress_bar(total=len(timesteps) - 1) as progress_bar:
             for t_curr, t_prev in zip(timesteps[:-1], timesteps[1:]):
-
                 # t_vec = torch.full((packed_latents.shape[0],), t_curr, dtype=packed_latents.dtype,
                 #                    device=device)
                 timestep = t_curr.expand(packed_latents.shape[0]).to(packed_latents.dtype)

From 8358ecd7de761acbac0b6d5b2937d038efc84c16 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Thu, 31 Oct 2024 14:29:46 +0200
Subject: [PATCH 05/45] fix import

---
 examples/community/pipeline_flux_rf_inversion.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index c3bf7a1def16..ee6c64d936d0 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -20,7 +20,7 @@
 from transformers import CLIPTextModel, CLIPTokenizer, T5EncoderModel, T5TokenizerFast
 
 from diffusers.image_processor import VaeImageProcessor
-from diffusers.loaders import FluxLoraLoaderMixin, FromSingleFileMixin
+from diffusers.loaders import FluxLoraLoaderMixin, FromSingleFileMixin, TextualInversionLoaderMixin
 from diffusers.models.autoencoders import AutoencoderKL
 from diffusers.models.transformers import FluxTransformer2DModel
 from diffusers.pipelines.flux.pipeline_output import FluxPipelineOutput

From 0d28b0e6078db8ab4d98f7a83045ed31112d1571 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Thu, 31 Oct 2024 14:43:08 +0200
Subject: [PATCH 06/45] fixes

---
 .../community/pipeline_flux_rf_inversion.py   | 29 ++++++++++---------
 1 file changed, 15 insertions(+), 14 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index ee6c64d936d0..3ecec1d4d8a5 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -19,7 +19,7 @@
 import torch
 from transformers import CLIPTextModel, CLIPTokenizer, T5EncoderModel, T5TokenizerFast
 
-from diffusers.image_processor import VaeImageProcessor
+from diffusers.image_processor import PipelineImageInput, VaeImageProcessor
 from diffusers.loaders import FluxLoraLoaderMixin, FromSingleFileMixin, TextualInversionLoaderMixin
 from diffusers.models.autoencoders import AutoencoderKL
 from diffusers.models.transformers import FluxTransformer2DModel
@@ -34,7 +34,7 @@
     scale_lora_layers,
     unscale_lora_layers,
 )
-
+from diffusers.utils.torch_utils import randn_tensor
 
 if is_torch_xla_available():
     import torch_xla.core.xla_model as xm
@@ -788,13 +788,13 @@ def __call__(
 
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:
-            for idx, (t_curr, t_prev) in enumerate(zip(timesteps[:-1], timesteps[1:])):
+            for i, (t_curr, t_prev) in enumerate(zip(timesteps[:-1], timesteps[1:])):
                 # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
                 # timestep = t.expand(latents.shape[0]).to(latents.dtype)
                 timestep = t_curr.expand(packed_inv_latents.shape[0]).to(packed_inv_latents.dtype)
 
                 velocity = self.transformer(
-                    hidden_states=latents,
+                    hidden_states=packed_inv_latents,
                     timestep=timestep / 1000,
                     guidance=guidance,
                     pooled_projections=pooled_prompt_embeds,
@@ -815,7 +815,7 @@ def __call__(
                 target_img_velocity = -(target_img - packed_inv_latents) / t_curr
 
                 # interpolated velocity
-                eta = eta_values[idx]
+                eta = eta_values[i]
                 interpolated_velocity = eta * target_img_velocity + (1 - eta) * velocity
                 latents = packed_inv_latents + (t_prev - t_curr) * interpolated_velocity
                 print(
@@ -824,16 +824,16 @@ def __call__(
 
                 # latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
 
-                if latents.dtype != latents_dtype:
-                    if torch.backends.mps.is_available():
-                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
-                        latents = latents.to(latents_dtype)
+                # if latents.dtype != latents_dtype:
+                #     if torch.backends.mps.is_available():
+                #         # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                #         latents = latents.to(latents_dtype)
 
                 if callback_on_step_end is not None:
                     callback_kwargs = {}
                     for k in callback_on_step_end_tensor_inputs:
                         callback_kwargs[k] = locals()[k]
-                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+                    callback_outputs = callback_on_step_end(self, i, t_curr, callback_kwargs)
 
                     latents = callback_outputs.pop("latents", latents)
                     prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
@@ -901,7 +901,8 @@ def invert(
         num_inversion_steps: int = 28,
         gamma: float = 0.5,
         use_shift_t_sampling: bool = False,
-        generator: Optional[torch.Generator] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
         dtype: Optional[torch.dtype] = None,
     ):
         dtype = dtype or self.text_encoder.dtype
@@ -935,7 +936,7 @@ def invert(
 
         timesteps = timesteps[::-1]  # flip for inversion
 
-        prompt_embeds, pooled_prompt_embeds, text_ids = pipeline.encode_prompt(prompt=source_prompt)
+        prompt_embeds, pooled_prompt_embeds, text_ids = self.encode_prompt(prompt=source_prompt)
         latent_image_ids = self._prepare_latent_image_ids(
             img_latents.shape[0],
             img_latents.shape[2],
@@ -968,7 +969,7 @@ def invert(
                 timestep = t_curr.expand(packed_latents.shape[0]).to(packed_latents.dtype)
 
                 # Null text velocity
-                velocity = pipeline.transformer(
+                velocity = self.transformer(
                     hidden_states=packed_latents,
                     timestep=timestep,
                     guidance=guidance,
@@ -977,7 +978,7 @@ def invert(
                     txt_ids=text_ids,
                     img_ids=latent_image_ids,
                     joint_attention_kwargs=None,
-                    return_dict=pipeline,
+                    return_dict=False,
                 )[0]
 
                 # Prevents precision issues

From f5df0304f50feee3620f2f6dfbda7b310d93dad2 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Thu, 31 Oct 2024 14:45:18 +0200
Subject: [PATCH 07/45] fixes

---
 examples/community/pipeline_flux_rf_inversion.py | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 3ecec1d4d8a5..20906250ea98 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -936,7 +936,7 @@ def invert(
 
         timesteps = timesteps[::-1]  # flip for inversion
 
-        prompt_embeds, pooled_prompt_embeds, text_ids = self.encode_prompt(prompt=source_prompt)
+        prompt_embeds, pooled_prompt_embeds, text_ids = self.encode_prompt(prompt=source_prompt, prompt_2=source_prompt)
         latent_image_ids = self._prepare_latent_image_ids(
             img_latents.shape[0],
             img_latents.shape[2],
@@ -945,11 +945,11 @@ def invert(
             dtype,
         )
         packed_latents = self._pack_latents(
-            latents,
-            batch_size=latents.shape[0],
-            num_channels_latents=latents.shape[1],
-            height=latents.shape[2],
-            width=latents.shape[3],
+            img_latents,
+            batch_size=img_latents.shape[0],
+            num_channels_latents=img_latents.shape[1],
+            height=img_latents.shape[2],
+            width=img_latents.shape[3],
         )
 
         target_noise = torch.randn(packed_latents.shape, device=device, dtype=torch.float32)

From ad66f17d2d65b6155f0b0a342996bcdd776bdbc6 Mon Sep 17 00:00:00 2001
From: Linoy <linoy@huggingface.co>
Date: Thu, 31 Oct 2024 12:48:42 +0000
Subject: [PATCH 08/45] style

---
 examples/community/pipeline_flux_rf_inversion.py | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 20906250ea98..0d19b29f1bae 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -36,6 +36,7 @@
 )
 from diffusers.utils.torch_utils import randn_tensor
 
+
 if is_torch_xla_available():
     import torch_xla.core.xla_model as xm
 
@@ -936,7 +937,9 @@ def invert(
 
         timesteps = timesteps[::-1]  # flip for inversion
 
-        prompt_embeds, pooled_prompt_embeds, text_ids = self.encode_prompt(prompt=source_prompt, prompt_2=source_prompt)
+        prompt_embeds, pooled_prompt_embeds, text_ids = self.encode_prompt(
+            prompt=source_prompt, prompt_2=source_prompt
+        )
         latent_image_ids = self._prepare_latent_image_ids(
             img_latents.shape[0],
             img_latents.shape[2],

From 757f98bd93a616e586f3912fa5047339d908a09c Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Fri, 1 Nov 2024 09:04:48 +0200
Subject: [PATCH 09/45] fixes

---
 examples/community/pipeline_flux_rf_inversion.py | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 20906250ea98..934cd8b17b7d 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -934,9 +934,11 @@ def invert(
             mu=mu,
         )
 
-        timesteps = timesteps[::-1]  # flip for inversion
+        timesteps = reversed(timesteps[-num_inversion_steps:])  # flip for inversion
 
-        prompt_embeds, pooled_prompt_embeds, text_ids = self.encode_prompt(prompt=source_prompt, prompt_2=source_prompt)
+        prompt_embeds, pooled_prompt_embeds, text_ids = self.encode_prompt(
+            prompt=source_prompt, prompt_2=source_prompt
+        )
         latent_image_ids = self._prepare_latent_image_ids(
             img_latents.shape[0],
             img_latents.shape[2],

From d2c4f7d97d1b3a152aa5802792cafc45e051059f Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Sun, 3 Nov 2024 15:37:43 +0200
Subject: [PATCH 10/45] change invert

---
 .../community/pipeline_flux_rf_inversion.py   | 176 +++++++++---------
 1 file changed, 87 insertions(+), 89 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index c084e52441b3..f764ee81ff3b 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -136,7 +136,6 @@ def retrieve_timesteps(
         timesteps = scheduler.timesteps
     return timesteps, num_inference_steps
 
-
 class RFInversionFluxPipeline(
     DiffusionPipeline,
     FluxLoraLoaderMixin,
@@ -565,6 +564,18 @@ def prepare_inverted_latents(
         latents = self._pack_latents(self.inverted_latents, batch_size, num_channels_latents, height, width)
         return latents, latent_image_ids
 
+    # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img.StableDiffusion3Img2ImgPipeline.get_timesteps
+    def get_timesteps(self, num_inference_steps, strength, device):
+        # get the original timestep using init_timestep
+        init_timestep = min(num_inference_steps * strength, num_inference_steps)
+
+        t_start = int(max(num_inference_steps - init_timestep, 0))
+        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order:]
+        if hasattr(self.scheduler, "set_begin_index"):
+            self.scheduler.set_begin_index(t_start * self.scheduler.order)
+
+        return timesteps, num_inference_steps - t_start
+
     @property
     def guidance_scale(self):
         return self._guidance_scale
@@ -901,114 +912,101 @@ def invert(
         source_guidance_scale=0.0,
         num_inversion_steps: int = 28,
         gamma: float = 0.5,
-        use_shift_t_sampling: bool = False,
         height: Optional[int] = None,
         width: Optional[int] = None,
+        timesteps: List[int] = None,
         dtype: Optional[torch.dtype] = None,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
     ):
         dtype = dtype or self.text_encoder.dtype
+        batch_size = 1
+        num_channels_latents = self.transformer.config.in_channels // 4
         height = height or self.default_sample_size * self.vae_scale_factor
         width = width or self.default_sample_size * self.vae_scale_factor
         device = self._execution_device
 
         # 1. prepare image
-        img_latents, _ = self.encode_image(image, dtype=dtype)
-
-        # 2. Prepare timesteps
+        image_latents, _ = self.encode_image(image, dtype=dtype)
+        _, latent_image_ids = self.prepare_latents(batch_size,
+                                                   num_channels_latents,
+                                                   height, width,
+                                                   dtype, device, generator)
+        image_latents = self._pack_latents(image_latents, batch_size, num_channels_latents, height, width)
+        self.image_latents = image_latents.clone()
+
+        # 2. prepare timesteps
         sigmas = np.linspace(1.0, 1 / num_inversion_steps, num_inversion_steps)
-        image_seq_len = img_latents.shape[1]
-        if use_shift_t_sampling:
-            mu = calculate_shift(
-                image_seq_len,
-                self.scheduler.config.base_image_seq_len,
-                self.scheduler.config.max_image_seq_len,
-                self.scheduler.config.base_shift,
-                self.scheduler.config.max_shift,
-            )
-        else:
-            mu = None
-        timesteps, num_inversion_steps = retrieve_timesteps(
+        image_seq_len = (int(height) // self.vae_scale_factor) * (int(width) // self.vae_scale_factor)
+        mu = calculate_shift(
+            image_seq_len,
+            self.scheduler.config.base_image_seq_len,
+            self.scheduler.config.max_image_seq_len,
+            self.scheduler.config.base_shift,
+            self.scheduler.config.max_shift,
+        )
+        timesteps, num_inference_steps = retrieve_timesteps(
             self.scheduler,
             num_inversion_steps,
             device,
-            sigmas=sigmas,
+            timesteps,
+            sigmas,
             mu=mu,
         )
+        timesteps, sigmas, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
 
-        timesteps = reversed(timesteps[-num_inversion_steps:])  # flip for inversion
-
-        prompt_embeds, pooled_prompt_embeds, text_ids = self.encode_prompt(
-            prompt=source_prompt, prompt_2=source_prompt
-        )
-        latent_image_ids = self._prepare_latent_image_ids(
-            img_latents.shape[0],
-            img_latents.shape[2],
-            img_latents.shape[3],
-            device,
-            dtype,
-        )
-        packed_latents = self._pack_latents(
-            img_latents,
-            batch_size=img_latents.shape[0],
-            num_channels_latents=img_latents.shape[1],
-            height=img_latents.shape[2],
-            width=img_latents.shape[3],
+        # 3. prepare text embeddings
+        (
+            prompt_embeds,
+            pooled_prompt_embeds,
+            text_ids,
+        ) = self.encode_prompt(
+            prompt=source_prompt,
+            prompt_2=source_prompt,
+            device=device,
         )
-
-        target_noise = torch.randn(packed_latents.shape, device=device, dtype=torch.float32)
-
-        # handle guidance
+        # 4. handle guidance
         if self.transformer.config.guidance_embeds:
             guidance = torch.full([1], source_guidance_scale, device=device, dtype=torch.float32)
-            guidance = guidance.expand(packed_latents.shape[0])
         else:
             guidance = None
 
-        # Image inversion with interpolated velocity field.  t goes from 0.0 to 1.0
-        with self.progress_bar(total=len(timesteps) - 1) as progress_bar:
-            for t_curr, t_prev in zip(timesteps[:-1], timesteps[1:]):
-                # t_vec = torch.full((packed_latents.shape[0],), t_curr, dtype=packed_latents.dtype,
-                #                    device=device)
-                timestep = t_curr.expand(packed_latents.shape[0]).to(packed_latents.dtype)
-
-                # Null text velocity
-                velocity = self.transformer(
-                    hidden_states=packed_latents,
-                    timestep=timestep,
-                    guidance=guidance,
-                    pooled_projections=pooled_prompt_embeds,
-                    encoder_hidden_states=prompt_embeds,
-                    txt_ids=text_ids,
-                    img_ids=latent_image_ids,
-                    joint_attention_kwargs=None,
-                    return_dict=False,
-                )[0]
-
-                # Prevents precision issues
-                packed_latents = packed_latents.to(torch.float32)
-                velocity = velocity.to(torch.float32)
-
-                # Target noise velocity
-                target_noise_velocity = (target_noise - packed_latents) / (1.0 - t_curr)
-
-                # interpolated velocity
-                interpolated_velocity = gamma * target_noise_velocity + (1 - gamma) * velocity
-
-                # one-step Euler
-                packed_latents = packed_latents + (t_prev - t_curr) * interpolated_velocity
-
-                packed_latents = packed_latents.to(dtype)
-                progress_bar.update()
-
-        print("Mean Absolute Error", torch.mean(torch.abs(packed_latents - target_noise)))
-
-        latents = self._unpack_latents(
-            packed_latents,
-            height=height,
-            width=width,
-            vae_scale_factor=self.vae_scale_factor,
-        )
-        latents = latents.to(dtype)
-        self.inverted_latents = latents
-        self.image_latents = img_latents
-        return latents
+        # if num_inference_steps < 1:
+        #     raise ValueError(
+        #         f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline"
+        #         f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline."
+        #     )
+        # latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
+
+        # Eq 8 dY_t = [u_t(Y_t) + γ(u_t(Y_t|y_1) - u_t(Y_t))]dt
+        Y_t = image_latents
+        y_1 = torch.randn_like(Y_t)
+        N = len(sigmas)
+
+        for i in range(N - 1):  # enumerate(timesteps):
+            t_i = torch.tensor(i / (N), dtype=Y_t.dtype, device=device)
+            timestep = torch.tensor(t_i, dtype=Y_t.dtype, device=device).repeat(batch_size)
+            # get the unconditional vector field
+
+            u_t_i = self.transformer(
+                hidden_states=Y_t,
+                timestep=timestep,
+                guidance=guidance,
+                pooled_projections=pooled_prompt_embeds,
+                encoder_hidden_states=prompt_embeds,
+                txt_ids=text_ids,
+                img_ids=latent_image_ids,
+                joint_attention_kwargs=self.joint_attention_kwargs,
+                return_dict=False,
+            )[0]
+
+            # get the conditional vector field
+            u_t_i_cond = (y_1 - Y_t) / (1 - t_i)
+
+            # controlled vector field
+            # Eq 8 dY_t = [u_t(Y_t) + γ(u_t(Y_t|y_1) - u_t(Y_t))]dt
+            u_hat_t_i = u_t_i + gamma * (u_t_i_cond - u_t_i)
+            Y_t = Y_t + u_hat_t_i * (sigmas[i] - sigmas[i+1])
+
+        self.inverted_latents = Y_t
+
+        return latent_image_ids

From ff75de8d6faaf4d5dce6a5a320a4784b5b1a6481 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Sun, 3 Nov 2024 16:00:21 +0200
Subject: [PATCH 11/45] change denoising & check inputs

---
 .../community/pipeline_flux_rf_inversion.py   | 102 +++++++++---------
 1 file changed, 49 insertions(+), 53 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index f764ee81ff3b..3fdb5a141ac9 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -401,10 +401,13 @@ def check_inputs(
         prompt_2,
         height,
         width,
+        start_timestep,
+        stop_timestep,
         prompt_embeds=None,
         pooled_prompt_embeds=None,
         callback_on_step_end_tensor_inputs=None,
         max_sequence_length=None,
+
     ):
         if height % self.vae_scale_factor != 0 or width % self.vae_scale_factor != 0:
             raise ValueError(
@@ -445,6 +448,10 @@ def check_inputs(
         if max_sequence_length is not None and max_sequence_length > 512:
             raise ValueError(f"`max_sequence_length` cannot be greater than 512 but is {max_sequence_length}")
 
+        # check start_timestep and stop_timestep
+        if start_timestep < 0 or start_timestep > stop_timestep:
+            raise ValueError(f"`start_timestep` should be in [0, stop_timestep] but is {start_timestep}")
+
     @staticmethod
     def _prepare_latent_image_ids(batch_size, height, width, device, dtype):
         latent_image_ids = torch.zeros(height, width, 3)
@@ -600,11 +607,12 @@ def __call__(
         prompt_2: Optional[Union[str, List[str]]] = None,
         height: Optional[int] = None,
         width: Optional[int] = None,
+        strength: float = 0.6,
+        eta: float = 1.0,
+        gamma: float = 1.0,
+        start_timestep: int = 0,
+        stop_timestep: int = 6,
         num_inference_steps: int = 28,
-        eta_base: float = 1.0,  # base eta value
-        eta_trend: str = "linear_decrease",  # constant, linear_increase, linear_decrease
-        start_step: int = 0,  # 0-based indexing, closed interval
-        end_step: int = 30,  # 0-based indexing, open interval
         timesteps: List[int] = None,
         use_shift_t_sampling: bool = True,
         guidance_scale: float = 3.5,
@@ -703,6 +711,7 @@ def __call__(
             pooled_prompt_embeds=pooled_prompt_embeds,
             callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
             max_sequence_length=max_sequence_length,
+
         )
 
         self._guidance_scale = guidance_scale
@@ -739,6 +748,10 @@ def __call__(
 
         # 4. Prepare latent variables
         num_channels_latents = self.transformer.config.in_channels // 4
+        latents = self.inverted_latents
+        latent_image_ids = self.latent_image_ids
+        image_latents = self.image_latents
+
         # latents, latent_image_ids = self.prepare_latents(
         #     batch_size * num_images_per_prompt,
         #     num_channels_latents,
@@ -749,28 +762,10 @@ def __call__(
         #     generator,
         #     latents,
         # )
-        packed_inv_latents, latent_image_ids = self.prepare_inverted_latents(
-            self.inverted_latents.shape[0],
-            self.inverted_latents.shape[1],
-            self.inverted_latents.shape[2],
-            self.inverted_latents.shape[3],
-            prompt_embeds.dtype,
-            device,
-            generator,
-        )
-
-        packed_img_latents = self._pack_latents(
-            self.image_latents,
-            batch_size=self.image_latents.shape[0],
-            num_channels_latents=self.image_latents.shape[1],
-            height=self.image_latents.shape[2],
-            width=self.image_latents.shape[3],
-        )
-        target_img = packed_img_latents.clone().to(torch.float32)
 
         # 5. Prepare timesteps
         sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
-        image_seq_len = packed_inv_latents.shape[1]
+        image_seq_len = latents.shape[1]
         mu = calculate_shift(
             image_seq_len,
             self.scheduler.config.base_image_seq_len,
@@ -792,21 +787,25 @@ def __call__(
         # handle guidance
         if self.transformer.config.guidance_embeds:
             guidance = torch.full([1], guidance_scale, device=device, dtype=torch.float32)
-            guidance = guidance.expand(packed_inv_latents.shape[0])
+            guidance = guidance.expand(latents.shape[0])
         else:
             guidance = None
 
-        eta_values = self.generate_eta_values(timesteps, start_step, end_step, eta_base, eta_trend)
-
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:
-            for i, (t_curr, t_prev) in enumerate(zip(timesteps[:-1], timesteps[1:])):
+            y_0 = image_latents.clone()
+            for i, t in enumerate(timesteps):
+                t_i = 1 - t / 1000  # torch.tensor((i+1) / (len(timesteps)-1), device=device)
+                # print(t_i, t)
+                dt = torch.tensor(1 / (len(timesteps) - 1), device=device)
+                if self.interrupt:
+                    continue
+
                 # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
-                # timestep = t.expand(latents.shape[0]).to(latents.dtype)
-                timestep = t_curr.expand(packed_inv_latents.shape[0]).to(packed_inv_latents.dtype)
+                timestep = t.expand(latents.shape[0]).to(latents.dtype)
 
-                velocity = self.transformer(
-                    hidden_states=packed_inv_latents,
+                v_t = -self.transformer(
+                    hidden_states=latents,
                     timestep=timestep / 1000,
                     guidance=guidance,
                     pooled_projections=pooled_prompt_embeds,
@@ -817,35 +816,31 @@ def __call__(
                     return_dict=False,
                 )[0]
 
-                # compute the previous noisy sample x_t -> x_t-1
-                # latents_dtype = latents.dtype
-                # Prevents precision issues
-                packed_inv_latents = packed_inv_latents.to(torch.float32)
-                velocity = velocity.to(torch.float32)
-
-                # Target image velocity
-                target_img_velocity = -(target_img - packed_inv_latents) / t_curr
+                v_t_cond = (y_0 - latents) / (1 - t_i)
+                eta_t = eta if start_timestep <= i < stop_timestep else 0.0
+                if start_timestep <= i < stop_timestep:
+                    # controlled vector field
+                    v_hat_t = v_t + eta * (v_t_cond - v_t)
+                    # v_hat_t = ((1 - t_i - eta_t) * latents  + eta_t * t_i * y_0) / (t_i*(1 - t_i)) + 2*(1-t_i)*(1 - eta_t) /t_i * v_t
 
-                # interpolated velocity
-                eta = eta_values[i]
-                interpolated_velocity = eta * target_img_velocity + (1 - eta) * velocity
-                latents = packed_inv_latents + (t_prev - t_curr) * interpolated_velocity
-                print(
-                    f"X_{t_prev:.3f} = X_{t_curr:.3f} + {t_prev - t_curr:.3f} * ({eta:.3f} * target_img_velocity + {1 - eta:.3f} * flux_velocity)"
-                )
+                else:
+                    v_hat_t = v_t
+                # SDE Eq: 17
 
-                # latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+                # compute the previous noisy sample x_t -> x_t-1
+                latents_dtype = latents.dtype
+                latents = latents + v_hat_t * (sigmas[i] - sigmas[i + 1])
 
-                # if latents.dtype != latents_dtype:
-                #     if torch.backends.mps.is_available():
-                #         # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
-                #         latents = latents.to(latents_dtype)
+                if latents.dtype != latents_dtype:
+                    if torch.backends.mps.is_available():
+                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                        latents = latents.to(latents_dtype)
 
                 if callback_on_step_end is not None:
                     callback_kwargs = {}
                     for k in callback_on_step_end_tensor_inputs:
                         callback_kwargs[k] = locals()[k]
-                    callback_outputs = callback_on_step_end(self, i, t_curr, callback_kwargs)
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
 
                     latents = callback_outputs.pop("latents", latents)
                     prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
@@ -1008,5 +1003,6 @@ def invert(
             Y_t = Y_t + u_hat_t_i * (sigmas[i] - sigmas[i+1])
 
         self.inverted_latents = Y_t
+        self.latent_image_ids = latent_image_ids
 
-        return latent_image_ids
+        return self.image_latents, Y_t, latent_image_ids

From f80ca4271cf90d96f175a89842304484e3f4a65c Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Sun, 3 Nov 2024 22:24:44 +0200
Subject: [PATCH 12/45] shape & timesteps fixes

---
 examples/community/pipeline_flux_rf_inversion.py | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 3fdb5a141ac9..839effe5ed79 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -577,11 +577,12 @@ def get_timesteps(self, num_inference_steps, strength, device):
         init_timestep = min(num_inference_steps * strength, num_inference_steps)
 
         t_start = int(max(num_inference_steps - init_timestep, 0))
-        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order:]
+        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+        sigmas = self.scheduler.sigmas[t_start * self.scheduler.order :]
         if hasattr(self.scheduler, "set_begin_index"):
             self.scheduler.set_begin_index(t_start * self.scheduler.order)
 
-        return timesteps, num_inference_steps - t_start
+        return timesteps, sigmas, num_inference_steps - t_start
 
     @property
     def guidance_scale(self):
@@ -707,6 +708,8 @@ def __call__(
             prompt_2,
             height,
             width,
+            start_timestep,
+            stop_timestep,
             prompt_embeds=prompt_embeds,
             pooled_prompt_embeds=pooled_prompt_embeds,
             callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
@@ -907,25 +910,32 @@ def invert(
         source_guidance_scale=0.0,
         num_inversion_steps: int = 28,
         gamma: float = 0.5,
+        strength: float = 0.6,
         height: Optional[int] = None,
         width: Optional[int] = None,
         timesteps: List[int] = None,
         dtype: Optional[torch.dtype] = None,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
     ):
         dtype = dtype or self.text_encoder.dtype
         batch_size = 1
+        self._joint_attention_kwargs = joint_attention_kwargs
         num_channels_latents = self.transformer.config.in_channels // 4
+
         height = height or self.default_sample_size * self.vae_scale_factor
         width = width or self.default_sample_size * self.vae_scale_factor
         device = self._execution_device
 
         # 1. prepare image
-        image_latents, _ = self.encode_image(image, dtype=dtype)
+        image_latents, _ = self.encode_image(image, height=height, width=width, dtype=dtype)
         _, latent_image_ids = self.prepare_latents(batch_size,
                                                    num_channels_latents,
                                                    height, width,
                                                    dtype, device, generator)
+
+        height = int(height) // self.vae_scale_factor
+        width = int(width) // self.vae_scale_factor
         image_latents = self._pack_latents(image_latents, batch_size, num_channels_latents, height, width)
         self.image_latents = image_latents.clone()
 

From d964300fccf6316407b2457f8198e396a1eb8232 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Mon, 4 Nov 2024 00:10:51 +0200
Subject: [PATCH 13/45] timesteps fixes

---
 examples/community/pipeline_flux_rf_inversion.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 839effe5ed79..773ffc040bae 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -784,6 +784,7 @@ def __call__(
             sigmas,
             mu=mu,
         )
+        timesteps, sigmas, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
         num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
         self._num_timesteps = len(timesteps)
 
@@ -797,6 +798,7 @@ def __call__(
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:
             y_0 = image_latents.clone()
+            print(f"y_0 shape: {y_0.shape}")
             for i, t in enumerate(timesteps):
                 t_i = 1 - t / 1000  # torch.tensor((i+1) / (len(timesteps)-1), device=device)
                 # print(t_i, t)

From 5834e7b25f95cc1fac9627f3eee1a6128a4c0e27 Mon Sep 17 00:00:00 2001
From: Linoy <linoy@huggingface.co>
Date: Sun, 3 Nov 2024 22:16:20 +0000
Subject: [PATCH 14/45] style

---
 examples/community/pipeline_flux_rf_inversion.py | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 773ffc040bae..06ef5588d08b 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -136,6 +136,7 @@ def retrieve_timesteps(
         timesteps = scheduler.timesteps
     return timesteps, num_inference_steps
 
+
 class RFInversionFluxPipeline(
     DiffusionPipeline,
     FluxLoraLoaderMixin,
@@ -407,7 +408,6 @@ def check_inputs(
         pooled_prompt_embeds=None,
         callback_on_step_end_tensor_inputs=None,
         max_sequence_length=None,
-
     ):
         if height % self.vae_scale_factor != 0 or width % self.vae_scale_factor != 0:
             raise ValueError(
@@ -714,7 +714,6 @@ def __call__(
             pooled_prompt_embeds=pooled_prompt_embeds,
             callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
             max_sequence_length=max_sequence_length,
-
         )
 
         self._guidance_scale = guidance_scale
@@ -931,10 +930,9 @@ def invert(
 
         # 1. prepare image
         image_latents, _ = self.encode_image(image, height=height, width=width, dtype=dtype)
-        _, latent_image_ids = self.prepare_latents(batch_size,
-                                                   num_channels_latents,
-                                                   height, width,
-                                                   dtype, device, generator)
+        _, latent_image_ids = self.prepare_latents(
+            batch_size, num_channels_latents, height, width, dtype, device, generator
+        )
 
         height = int(height) // self.vae_scale_factor
         width = int(width) // self.vae_scale_factor
@@ -1012,7 +1010,7 @@ def invert(
             # controlled vector field
             # Eq 8 dY_t = [u_t(Y_t) + γ(u_t(Y_t|y_1) - u_t(Y_t))]dt
             u_hat_t_i = u_t_i + gamma * (u_t_i_cond - u_t_i)
-            Y_t = Y_t + u_hat_t_i * (sigmas[i] - sigmas[i+1])
+            Y_t = Y_t + u_hat_t_i * (sigmas[i] - sigmas[i + 1])
 
         self.inverted_latents = Y_t
         self.latent_image_ids = latent_image_ids

From 910bce7176521d5ae342bbbe1f8c3b1a4b558925 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Mon, 4 Nov 2024 10:45:55 +0200
Subject: [PATCH 15/45] remove redundancies

---
 .../community/pipeline_flux_rf_inversion.py   | 40 +------------------
 1 file changed, 1 insertion(+), 39 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 773ffc040bae..ebfecd38c172 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -615,7 +615,6 @@ def __call__(
         stop_timestep: int = 6,
         num_inference_steps: int = 28,
         timesteps: List[int] = None,
-        use_shift_t_sampling: bool = True,
         guidance_scale: float = 3.5,
         num_images_per_prompt: Optional[int] = 1,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
@@ -874,36 +873,6 @@ def __call__(
 
         return FluxPipelineOutput(images=image)
 
-    def generate_eta_values(
-        self,
-        timesteps,
-        start_step,
-        end_step,
-        eta,
-        eta_trend,
-    ):
-        assert (
-            start_step < end_step and start_step >= 0 and end_step <= len(timesteps)
-        ), "Invalid start_step and end_step"
-        # timesteps are monotonically decreasing, from 1.0 to 0.0
-        eta_values = [0.0] * (len(timesteps) - 1)
-
-        if eta_trend == "constant":
-            for i in range(start_step, end_step):
-                eta_values[i] = eta
-        elif eta_trend == "linear_increase":
-            total_time = timesteps[start_step] - timesteps[end_step - 1]
-            for i in range(start_step, end_step):
-                eta_values[i] = eta * (timesteps[start_step] - timesteps[i]) / total_time
-        elif eta_trend == "linear_decrease":
-            total_time = timesteps[start_step] - timesteps[end_step - 1]
-            for i in range(start_step, end_step):
-                eta_values[i] = eta * (timesteps[i] - timesteps[end_step - 1]) / total_time
-        else:
-            raise NotImplementedError(f"Unsupported eta_trend: {eta_trend}")
-
-        return eta_values
-
     @torch.no_grad()
     def invert(
         self,
@@ -977,13 +946,6 @@ def invert(
         else:
             guidance = None
 
-        # if num_inference_steps < 1:
-        #     raise ValueError(
-        #         f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline"
-        #         f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline."
-        #     )
-        # latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
-
         # Eq 8 dY_t = [u_t(Y_t) + γ(u_t(Y_t|y_1) - u_t(Y_t))]dt
         Y_t = image_latents
         y_1 = torch.randn_like(Y_t)
@@ -992,8 +954,8 @@ def invert(
         for i in range(N - 1):  # enumerate(timesteps):
             t_i = torch.tensor(i / (N), dtype=Y_t.dtype, device=device)
             timestep = torch.tensor(t_i, dtype=Y_t.dtype, device=device).repeat(batch_size)
-            # get the unconditional vector field
 
+            # get the unconditional vector field
             u_t_i = self.transformer(
                 hidden_states=Y_t,
                 timestep=timestep,

From 156e611669dc145be8e5c5d027fb07020f7d10a4 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Mon, 4 Nov 2024 11:18:24 +0200
Subject: [PATCH 16/45] small changes

---
 .../community/pipeline_flux_rf_inversion.py    | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index bdc7c3416d98..08b70c3428be 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -795,12 +795,13 @@ def __call__(
 
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:
+
             y_0 = image_latents.clone()
-            print(f"y_0 shape: {y_0.shape}")
             for i, t in enumerate(timesteps):
-                t_i = 1 - t / 1000  # torch.tensor((i+1) / (len(timesteps)-1), device=device)
-                # print(t_i, t)
+
+                t_i = 1 - t / 1000
                 dt = torch.tensor(1 / (len(timesteps) - 1), device=device)
+
                 if self.interrupt:
                     continue
 
@@ -824,13 +825,11 @@ def __call__(
                 if start_timestep <= i < stop_timestep:
                     # controlled vector field
                     v_hat_t = v_t + eta * (v_t_cond - v_t)
-                    # v_hat_t = ((1 - t_i - eta_t) * latents  + eta_t * t_i * y_0) / (t_i*(1 - t_i)) + 2*(1-t_i)*(1 - eta_t) /t_i * v_t
 
                 else:
                     v_hat_t = v_t
                 # SDE Eq: 17
 
-                # compute the previous noisy sample x_t -> x_t-1
                 latents_dtype = latents.dtype
                 latents = latents + v_hat_t * (sigmas[i] - sigmas[i + 1])
 
@@ -918,7 +917,7 @@ def invert(
             self.scheduler.config.base_shift,
             self.scheduler.config.max_shift,
         )
-        timesteps, num_inference_steps = retrieve_timesteps(
+        timesteps, num_inversion_steps = retrieve_timesteps(
             self.scheduler,
             num_inversion_steps,
             device,
@@ -926,7 +925,7 @@ def invert(
             sigmas,
             mu=mu,
         )
-        timesteps, sigmas, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
+        timesteps, sigmas, num_inversion_steps = self.get_timesteps(num_inversion_steps, strength, device)
 
         # 3. prepare text embeddings
         (
@@ -947,10 +946,9 @@ def invert(
         # Eq 8 dY_t = [u_t(Y_t) + γ(u_t(Y_t|y_1) - u_t(Y_t))]dt
         Y_t = image_latents
         y_1 = torch.randn_like(Y_t)
-        N = len(sigmas)
 
-        for i in range(N - 1):  # enumerate(timesteps):
-            t_i = torch.tensor(i / (N), dtype=Y_t.dtype, device=device)
+        for i in range(num_inversion_steps - 1):
+            t_i = torch.tensor(i / (num_inversion_steps), dtype=Y_t.dtype, device=device)
             timestep = torch.tensor(t_i, dtype=Y_t.dtype, device=device).repeat(batch_size)
 
             # get the unconditional vector field

From b32a08cf83609a4afa974ff7e0898ec4ab9b561b Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Mon, 4 Nov 2024 14:16:45 +0200
Subject: [PATCH 17/45] update documentation a bit

---
 .../community/pipeline_flux_rf_inversion.py   | 37 +++++++++++++++++--
 1 file changed, 33 insertions(+), 4 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 08b70c3428be..debfe5de7420 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -572,7 +572,7 @@ def prepare_inverted_latents(
         return latents, latent_image_ids
 
     # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img.StableDiffusion3Img2ImgPipeline.get_timesteps
-    def get_timesteps(self, num_inference_steps, strength, device):
+    def get_timesteps(self, num_inference_steps, strength):
         # get the original timestep using init_timestep
         init_timestep = min(num_inference_steps * strength, num_inference_steps)
 
@@ -610,7 +610,6 @@ def __call__(
         width: Optional[int] = None,
         strength: float = 0.6,
         eta: float = 1.0,
-        gamma: float = 1.0,
         start_timestep: int = 0,
         stop_timestep: int = 6,
         num_inference_steps: int = 28,
@@ -642,6 +641,9 @@ def __call__(
                 The height in pixels of the generated image. This is set to 1024 by default for the best results.
             width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                 The width in pixels of the generated image. This is set to 1024 by default for the best results.
+            eta (`float`, *optional*, defaults to 1.0):
+                The controller guidance, balancing faithfulness & editability:
+                higher eta - better faithfullness, less editability. For more significant edits, lower the value of eta.
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                 expense of slower inference.
@@ -782,7 +784,7 @@ def __call__(
             sigmas,
             mu=mu,
         )
-        timesteps, sigmas, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
+        timesteps, sigmas, num_inference_steps = self.get_timesteps(num_inference_steps, strength)
         num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
         self._num_timesteps = len(timesteps)
 
@@ -887,6 +889,33 @@ def invert(
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
         joint_attention_kwargs: Optional[Dict[str, Any]] = None,
     ):
+        r"""
+        Performs Algorithm 1: Controlled Forward ODE from https://arxiv.org/pdf/2410.10792
+        Args:
+            image (`PipelineImageInput`):
+                Input for the image(s) that are to be edited. Multiple input images have to default to the same aspect
+                ratio.
+            source_prompt (`str` or `List[str]`, *optional* defaults to an empty prompt as done in the original paper):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            source_guidance_scale (`float`, *optional*, defaults to 0.0):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). For this algorithm, it's better to keep it 0.
+            num_inversion_steps (`int`, *optional*, defaults to 28):
+                The number of discretization steps.
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The height in pixels of the generated image. This is set to 1024 by default for the best results.
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The width in pixels of the generated image. This is set to 1024 by default for the best results.
+            gamma (`float`, *optional*, defaults to 0.5):
+                The controller guidance for the forward ODE, balancing faithfulness & editability:
+                higher eta - better faithfullness, less editability. For more significant edits, lower the value of eta.
+            timesteps (`List[int]`, *optional*):
+                Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+                in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+                passed will be used. Must be in descending order.
+        """
         dtype = dtype or self.text_encoder.dtype
         batch_size = 1
         self._joint_attention_kwargs = joint_attention_kwargs
@@ -925,7 +954,7 @@ def invert(
             sigmas,
             mu=mu,
         )
-        timesteps, sigmas, num_inversion_steps = self.get_timesteps(num_inversion_steps, strength, device)
+        timesteps, sigmas, num_inversion_steps = self.get_timesteps(num_inversion_steps, strength)
 
         # 3. prepare text embeddings
         (

From 87146476c7be56ba50ccecc847b4c4a2601459f0 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Mon, 4 Nov 2024 14:24:43 +0200
Subject: [PATCH 18/45] update documentation a bit

---
 .../community/pipeline_flux_rf_inversion.py   | 19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index debfe5de7420..d818e5926a57 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -55,11 +55,20 @@
 
         >>> pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", torch_dtype=torch.bfloat16)
         >>> pipe.to("cuda")
-        >>> prompt = "A cat holding a sign that says hello world"
-        >>> # Depending on the variant being used, the pipeline call will slightly vary.
-        >>> # Refer to the pipeline documentation for more details.
-        >>> image = pipe(prompt, num_inference_steps=4, guidance_scale=0.0).images[0]
-        >>> image.save("flux.png")
+        
+         >>> def download_image(url):
+        ...     response = requests.get(url)
+        ...     return PIL.Image.open(BytesIO(response.content)).convert("RGB")
+
+
+        >>> img_url = "https://www.aiml.informatik.tu-darmstadt.de/people/mbrack/tennis.jpg"
+        >>> image = download_image(img_url)
+
+        >>> _,__,___ = pipe.invert(image=image, num_inversion_steps=28)
+
+        >>> edited_image = pipe(
+        ...     prompt="a portrait of a tiger",
+        ... ).images[0]
         ```
 """
 

From c64a0e72610b5745b0d0dd194b12f21f91fad5d2 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Mon, 4 Nov 2024 15:31:06 +0200
Subject: [PATCH 19/45] update documentation a bit

---
 examples/community/pipeline_flux_rf_inversion.py | 11 -----------
 1 file changed, 11 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index d818e5926a57..75a1124e3d90 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -764,17 +764,6 @@ def __call__(
         latent_image_ids = self.latent_image_ids
         image_latents = self.image_latents
 
-        # latents, latent_image_ids = self.prepare_latents(
-        #     batch_size * num_images_per_prompt,
-        #     num_channels_latents,
-        #     height,
-        #     width,
-        #     prompt_embeds.dtype,
-        #     device,
-        #     generator,
-        #     latents,
-        # )
-
         # 5. Prepare timesteps
         sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
         image_seq_len = latents.shape[1]

From 142aef569c1d0efef405b3178924e23de2661114 Mon Sep 17 00:00:00 2001
From: Linoy <linoy@huggingface.co>
Date: Mon, 4 Nov 2024 13:32:22 +0000
Subject: [PATCH 20/45] style

---
 examples/community/pipeline_flux_rf_inversion.py | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 75a1124e3d90..2a757c97e8c2 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -55,7 +55,7 @@
 
         >>> pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", torch_dtype=torch.bfloat16)
         >>> pipe.to("cuda")
-        
+
          >>> def download_image(url):
         ...     response = requests.get(url)
         ...     return PIL.Image.open(BytesIO(response.content)).convert("RGB")
@@ -795,10 +795,8 @@ def __call__(
 
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:
-
             y_0 = image_latents.clone()
             for i, t in enumerate(timesteps):
-
                 t_i = 1 - t / 1000
                 dt = torch.tensor(1 / (len(timesteps) - 1), device=device)
 

From 9d8b37c10c8f02d15da708bf7c1a7cddc6d1bc0d Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Tue, 5 Nov 2024 15:30:56 +0200
Subject: [PATCH 21/45] change strength param, remove redundancies

---
 .../community/pipeline_flux_rf_inversion.py   | 104 ++++++++----------
 1 file changed, 43 insertions(+), 61 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 75a1124e3d90..4422d03fb7c6 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -559,31 +559,10 @@ def prepare_latents(
 
         return latents, latent_image_ids
 
-    def prepare_inverted_latents(
-        self,
-        batch_size,
-        num_channels_latents,
-        height,
-        width,
-        dtype,
-        device,
-        generator,
-    ):
-        latent_image_ids = self._prepare_latent_image_ids(batch_size, height, width, device, dtype)
-
-        if isinstance(generator, list) and len(generator) != batch_size:
-            raise ValueError(
-                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
-                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
-            )
-
-        latents = self._pack_latents(self.inverted_latents, batch_size, num_channels_latents, height, width)
-        return latents, latent_image_ids
-
     # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img.StableDiffusion3Img2ImgPipeline.get_timesteps
-    def get_timesteps(self, num_inference_steps, strength):
+    def get_timesteps(self, num_inference_steps, timestep_offset):
         # get the original timestep using init_timestep
-        init_timestep = min(num_inference_steps * strength, num_inference_steps)
+        init_timestep = min(num_inference_steps * timestep_offset, num_inference_steps)
 
         t_start = int(max(num_inference_steps - init_timestep, 0))
         timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
@@ -612,29 +591,29 @@ def interrupt(self):
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
-        self,
-        prompt: Union[str, List[str]] = None,
-        prompt_2: Optional[Union[str, List[str]]] = None,
-        height: Optional[int] = None,
-        width: Optional[int] = None,
-        strength: float = 0.6,
-        eta: float = 1.0,
-        start_timestep: int = 0,
-        stop_timestep: int = 6,
-        num_inference_steps: int = 28,
-        timesteps: List[int] = None,
-        guidance_scale: float = 3.5,
-        num_images_per_prompt: Optional[int] = 1,
-        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        latents: Optional[torch.FloatTensor] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
-        output_type: Optional[str] = "pil",
-        return_dict: bool = True,
-        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
-        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
-        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
-        max_sequence_length: int = 512,
+            self,
+            prompt: Union[str, List[str]] = None,
+            prompt_2: Optional[Union[str, List[str]]] = None,
+            height: Optional[int] = None,
+            width: Optional[int] = None,
+            eta: float = 1.0,
+            timestep_offset: float = 0.6,
+            start_timestep: float = 0.,
+            stop_timestep: float = 0.25,
+            num_inference_steps: int = 28,
+            timesteps: List[int] = None,
+            guidance_scale: float = 3.5,
+            num_images_per_prompt: Optional[int] = 1,
+            generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+            latents: Optional[torch.FloatTensor] = None,
+            prompt_embeds: Optional[torch.FloatTensor] = None,
+            pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
+            output_type: Optional[str] = "pil",
+            return_dict: bool = True,
+            joint_attention_kwargs: Optional[Dict[str, Any]] = None,
+            callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+            callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+            max_sequence_length: int = 512,
     ):
         r"""
         Function invoked when calling the pipeline for generation.
@@ -782,7 +761,10 @@ def __call__(
             sigmas,
             mu=mu,
         )
-        timesteps, sigmas, num_inference_steps = self.get_timesteps(num_inference_steps, strength)
+        start_timestep = int(start_timestep * num_inference_steps)
+        stop_timestep = min(int(stop_timestep * num_inference_steps), num_inference_steps)
+
+        timesteps, sigmas, num_inference_steps = self.get_timesteps(num_inference_steps, timestep_offset)
         num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
         self._num_timesteps = len(timesteps)
 
@@ -873,19 +855,19 @@ def __call__(
 
     @torch.no_grad()
     def invert(
-        self,
-        image: PipelineImageInput,
-        source_prompt: str = "",
-        source_guidance_scale=0.0,
-        num_inversion_steps: int = 28,
-        gamma: float = 0.5,
-        strength: float = 0.6,
-        height: Optional[int] = None,
-        width: Optional[int] = None,
-        timesteps: List[int] = None,
-        dtype: Optional[torch.dtype] = None,
-        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
+            self,
+            image: PipelineImageInput,
+            source_prompt: str = "",
+            source_guidance_scale=0.0,
+            num_inversion_steps: int = 28,
+            timestep_offset: float = 0.6,
+            gamma: float = 0.5,
+            height: Optional[int] = None,
+            width: Optional[int] = None,
+            timesteps: List[int] = None,
+            dtype: Optional[torch.dtype] = None,
+            generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+            joint_attention_kwargs: Optional[Dict[str, Any]] = None,
     ):
         r"""
         Performs Algorithm 1: Controlled Forward ODE from https://arxiv.org/pdf/2410.10792
@@ -952,7 +934,7 @@ def invert(
             sigmas,
             mu=mu,
         )
-        timesteps, sigmas, num_inversion_steps = self.get_timesteps(num_inversion_steps, strength)
+        timesteps, sigmas, num_inversion_steps = self.get_timesteps(num_inversion_steps, timestep_offset=timestep_offset)
 
         # 3. prepare text embeddings
         (

From 61aea50b7b4fe01dccda0ac3ea9f00ae2dbc2b91 Mon Sep 17 00:00:00 2001
From: Linoy <linoy@huggingface.co>
Date: Tue, 5 Nov 2024 13:52:44 +0000
Subject: [PATCH 22/45] style

---
 .../community/pipeline_flux_rf_inversion.py   | 76 ++++++++++---------
 1 file changed, 39 insertions(+), 37 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index ec8852783429..880c0a218fd7 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -591,29 +591,29 @@ def interrupt(self):
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
-            self,
-            prompt: Union[str, List[str]] = None,
-            prompt_2: Optional[Union[str, List[str]]] = None,
-            height: Optional[int] = None,
-            width: Optional[int] = None,
-            eta: float = 1.0,
-            timestep_offset: float = 0.6,
-            start_timestep: float = 0.,
-            stop_timestep: float = 0.25,
-            num_inference_steps: int = 28,
-            timesteps: List[int] = None,
-            guidance_scale: float = 3.5,
-            num_images_per_prompt: Optional[int] = 1,
-            generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-            latents: Optional[torch.FloatTensor] = None,
-            prompt_embeds: Optional[torch.FloatTensor] = None,
-            pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
-            output_type: Optional[str] = "pil",
-            return_dict: bool = True,
-            joint_attention_kwargs: Optional[Dict[str, Any]] = None,
-            callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
-            callback_on_step_end_tensor_inputs: List[str] = ["latents"],
-            max_sequence_length: int = 512,
+        self,
+        prompt: Union[str, List[str]] = None,
+        prompt_2: Optional[Union[str, List[str]]] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        eta: float = 1.0,
+        timestep_offset: float = 0.6,
+        start_timestep: float = 0.0,
+        stop_timestep: float = 0.25,
+        num_inference_steps: int = 28,
+        timesteps: List[int] = None,
+        guidance_scale: float = 3.5,
+        num_images_per_prompt: Optional[int] = 1,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
+        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        max_sequence_length: int = 512,
     ):
         r"""
         Function invoked when calling the pipeline for generation.
@@ -853,19 +853,19 @@ def __call__(
 
     @torch.no_grad()
     def invert(
-            self,
-            image: PipelineImageInput,
-            source_prompt: str = "",
-            source_guidance_scale=0.0,
-            num_inversion_steps: int = 28,
-            timestep_offset: float = 0.6,
-            gamma: float = 0.5,
-            height: Optional[int] = None,
-            width: Optional[int] = None,
-            timesteps: List[int] = None,
-            dtype: Optional[torch.dtype] = None,
-            generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-            joint_attention_kwargs: Optional[Dict[str, Any]] = None,
+        self,
+        image: PipelineImageInput,
+        source_prompt: str = "",
+        source_guidance_scale=0.0,
+        num_inversion_steps: int = 28,
+        timestep_offset: float = 0.6,
+        gamma: float = 0.5,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        timesteps: List[int] = None,
+        dtype: Optional[torch.dtype] = None,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
     ):
         r"""
         Performs Algorithm 1: Controlled Forward ODE from https://arxiv.org/pdf/2410.10792
@@ -932,7 +932,9 @@ def invert(
             sigmas,
             mu=mu,
         )
-        timesteps, sigmas, num_inversion_steps = self.get_timesteps(num_inversion_steps, timestep_offset=timestep_offset)
+        timesteps, sigmas, num_inversion_steps = self.get_timesteps(
+            num_inversion_steps, timestep_offset=timestep_offset
+        )
 
         # 3. prepare text embeddings
         (

From 61c810572b170d8573cd9349ebe4a61baecd2087 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Mon, 25 Nov 2024 10:27:22 +0200
Subject: [PATCH 23/45] forward ode loop change

---
 .../community/pipeline_flux_rf_inversion.py   | 51 ++++++++++---------
 1 file changed, 27 insertions(+), 24 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 880c0a218fd7..525110dc9ea8 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -955,31 +955,34 @@ def invert(
         # Eq 8 dY_t = [u_t(Y_t) + γ(u_t(Y_t|y_1) - u_t(Y_t))]dt
         Y_t = image_latents
         y_1 = torch.randn_like(Y_t)
+        N = len(sigmas)
 
-        for i in range(num_inversion_steps - 1):
-            t_i = torch.tensor(i / (num_inversion_steps), dtype=Y_t.dtype, device=device)
-            timestep = torch.tensor(t_i, dtype=Y_t.dtype, device=device).repeat(batch_size)
-
-            # get the unconditional vector field
-            u_t_i = self.transformer(
-                hidden_states=Y_t,
-                timestep=timestep,
-                guidance=guidance,
-                pooled_projections=pooled_prompt_embeds,
-                encoder_hidden_states=prompt_embeds,
-                txt_ids=text_ids,
-                img_ids=latent_image_ids,
-                joint_attention_kwargs=self.joint_attention_kwargs,
-                return_dict=False,
-            )[0]
-
-            # get the conditional vector field
-            u_t_i_cond = (y_1 - Y_t) / (1 - t_i)
-
-            # controlled vector field
-            # Eq 8 dY_t = [u_t(Y_t) + γ(u_t(Y_t|y_1) - u_t(Y_t))]dt
-            u_hat_t_i = u_t_i + gamma * (u_t_i_cond - u_t_i)
-            Y_t = Y_t + u_hat_t_i * (sigmas[i] - sigmas[i + 1])
+        # forward ODE loop
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i in range(N - 1):
+                t_i = torch.tensor(i / (N), dtype=Y_t.dtype, device=device)
+                timestep = torch.tensor(t_i, dtype=Y_t.dtype, device=device).repeat(batch_size)
+
+                # get the unconditional vector field
+                u_t_i = self.transformer(
+                    hidden_states=Y_t,
+                    timestep=timestep,
+                    guidance=guidance,
+                    pooled_projections=pooled_prompt_embeds,
+                    encoder_hidden_states=prompt_embeds,
+                    txt_ids=text_ids,
+                    img_ids=latent_image_ids,
+                    joint_attention_kwargs=self.joint_attention_kwargs,
+                    return_dict=False,
+                )[0]
+
+                # get the conditional vector field
+                u_t_i_cond = (y_1 - Y_t) / (1 - t_i)
+
+                # controlled vector field
+                # Eq 8 dY_t = [u_t(Y_t) + γ(u_t(Y_t|y_1) - u_t(Y_t))]dt
+                u_hat_t_i = u_t_i + gamma * (u_t_i_cond - u_t_i)
+                Y_t = Y_t + u_hat_t_i * (sigmas[i] - sigmas[i + 1])
 
         self.inverted_latents = Y_t
         self.latent_image_ids = latent_image_ids

From 093a91bddddeb432d938119f88cace75559fba3a Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Mon, 25 Nov 2024 14:09:56 +0200
Subject: [PATCH 24/45] add inversion progress bar

---
 examples/community/pipeline_flux_rf_inversion.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 525110dc9ea8..9a8b7ef08fa0 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -958,7 +958,7 @@ def invert(
         N = len(sigmas)
 
         # forward ODE loop
-        with self.progress_bar(total=num_inference_steps) as progress_bar:
+        with self.progress_bar(total=N) as progress_bar:
             for i in range(N - 1):
                 t_i = torch.tensor(i / (N), dtype=Y_t.dtype, device=device)
                 timestep = torch.tensor(t_i, dtype=Y_t.dtype, device=device).repeat(batch_size)
@@ -983,8 +983,9 @@ def invert(
                 # Eq 8 dY_t = [u_t(Y_t) + γ(u_t(Y_t|y_1) - u_t(Y_t))]dt
                 u_hat_t_i = u_t_i + gamma * (u_t_i_cond - u_t_i)
                 Y_t = Y_t + u_hat_t_i * (sigmas[i] - sigmas[i + 1])
+                progress_bar.update()
 
         self.inverted_latents = Y_t
         self.latent_image_ids = latent_image_ids
 
-        return self.image_latents, Y_t, latent_image_ids
+        return self.image_latents, Y_t, latent_image_ids
\ No newline at end of file

From e1c905c3a34f01e453794e0597b37bb902d7cf3b Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Mon, 25 Nov 2024 14:41:01 +0200
Subject: [PATCH 25/45] fix image_seq_len

---
 .../community/pipeline_flux_rf_inversion.py   | 32 ++++---------------
 1 file changed, 7 insertions(+), 25 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 9a8b7ef08fa0..0d6fc3557be8 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -534,26 +534,12 @@ def prepare_latents(
         width,
         dtype,
         device,
-        generator,
-        latents=None,
+        image_latents,
     ):
         height = int(height) // self.vae_scale_factor
         width = int(width) // self.vae_scale_factor
 
-        shape = (batch_size, num_channels_latents, height, width)
-
-        if latents is not None:
-            latent_image_ids = self._prepare_latent_image_ids(batch_size, height, width, device, dtype)
-            return latents.to(device=device, dtype=dtype), latent_image_ids
-
-        if isinstance(generator, list) and len(generator) != batch_size:
-            raise ValueError(
-                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
-                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
-            )
-
-        latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
-        latents = self._pack_latents(latents, batch_size, num_channels_latents, height, width)
+        latents = self._pack_latents(image_latents, batch_size, num_channels_latents, height, width)
 
         latent_image_ids = self._prepare_latent_image_ids(batch_size, height // 2, width // 2, device, dtype)
 
@@ -745,7 +731,7 @@ def __call__(
 
         # 5. Prepare timesteps
         sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
-        image_seq_len = latents.shape[1]
+        image_seq_len = (int(height) // self.vae_scale_factor // 2) * (int(width) // self.vae_scale_factor // 2)
         mu = calculate_shift(
             image_seq_len,
             self.scheduler.config.base_image_seq_len,
@@ -905,18 +891,14 @@ def invert(
 
         # 1. prepare image
         image_latents, _ = self.encode_image(image, height=height, width=width, dtype=dtype)
-        _, latent_image_ids = self.prepare_latents(
-            batch_size, num_channels_latents, height, width, dtype, device, generator
+        image_latents, latent_image_ids = self.prepare_latents(
+            batch_size, num_channels_latents, height, width, dtype, device, image_latents
         )
-
-        height = int(height) // self.vae_scale_factor
-        width = int(width) // self.vae_scale_factor
-        image_latents = self._pack_latents(image_latents, batch_size, num_channels_latents, height, width)
         self.image_latents = image_latents.clone()
 
         # 2. prepare timesteps
         sigmas = np.linspace(1.0, 1 / num_inversion_steps, num_inversion_steps)
-        image_seq_len = (int(height) // self.vae_scale_factor) * (int(width) // self.vae_scale_factor)
+        image_seq_len = (int(height) // self.vae_scale_factor // 2) * (int(width) // self.vae_scale_factor // 2)
         mu = calculate_shift(
             image_seq_len,
             self.scheduler.config.base_image_seq_len,
@@ -958,7 +940,7 @@ def invert(
         N = len(sigmas)
 
         # forward ODE loop
-        with self.progress_bar(total=N) as progress_bar:
+        with self.progress_bar(total=N-1) as progress_bar:
             for i in range(N - 1):
                 t_i = torch.tensor(i / (N), dtype=Y_t.dtype, device=device)
                 timestep = torch.tensor(t_i, dtype=Y_t.dtype, device=device).repeat(batch_size)

From 645885ff79af9c482a10fde22adafadb15f43971 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Mon, 25 Nov 2024 18:07:57 +0200
Subject: [PATCH 26/45] revert to strength but == 1 by default.

---
 examples/community/pipeline_flux_rf_inversion.py | 15 +++++++--------
 1 file changed, 7 insertions(+), 8 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 0d6fc3557be8..75e738e00947 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -546,9 +546,9 @@ def prepare_latents(
         return latents, latent_image_ids
 
     # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img.StableDiffusion3Img2ImgPipeline.get_timesteps
-    def get_timesteps(self, num_inference_steps, timestep_offset):
+    def get_timesteps(self, num_inference_steps, strength=1.):
         # get the original timestep using init_timestep
-        init_timestep = min(num_inference_steps * timestep_offset, num_inference_steps)
+        init_timestep = min(num_inference_steps * strength, num_inference_steps)
 
         t_start = int(max(num_inference_steps - init_timestep, 0))
         timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
@@ -583,8 +583,8 @@ def __call__(
         height: Optional[int] = None,
         width: Optional[int] = None,
         eta: float = 1.0,
-        timestep_offset: float = 0.6,
-        start_timestep: float = 0.0,
+        strength: float = 1.,
+        start_timestep: float = 0,
         stop_timestep: float = 0.25,
         num_inference_steps: int = 28,
         timesteps: List[int] = None,
@@ -749,8 +749,7 @@ def __call__(
         )
         start_timestep = int(start_timestep * num_inference_steps)
         stop_timestep = min(int(stop_timestep * num_inference_steps), num_inference_steps)
-
-        timesteps, sigmas, num_inference_steps = self.get_timesteps(num_inference_steps, timestep_offset)
+        timesteps, sigmas, num_inference_steps = self.get_timesteps(num_inference_steps, strength)
         num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
         self._num_timesteps = len(timesteps)
 
@@ -844,7 +843,7 @@ def invert(
         source_prompt: str = "",
         source_guidance_scale=0.0,
         num_inversion_steps: int = 28,
-        timestep_offset: float = 0.6,
+        strength: float = 1.,
         gamma: float = 0.5,
         height: Optional[int] = None,
         width: Optional[int] = None,
@@ -915,7 +914,7 @@ def invert(
             mu=mu,
         )
         timesteps, sigmas, num_inversion_steps = self.get_timesteps(
-            num_inversion_steps, timestep_offset=timestep_offset
+            num_inversion_steps, strength
         )
 
         # 3. prepare text embeddings

From d887572c8151b7f13ac2363796b2f6007fc99ade Mon Sep 17 00:00:00 2001
From: Linoy <linoy@huggingface.co>
Date: Mon, 25 Nov 2024 16:09:51 +0000
Subject: [PATCH 27/45] style

---
 examples/community/pipeline_flux_rf_inversion.py | 15 ++++++---------
 1 file changed, 6 insertions(+), 9 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 75e738e00947..77aa6d436ea0 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -34,7 +34,6 @@
     scale_lora_layers,
     unscale_lora_layers,
 )
-from diffusers.utils.torch_utils import randn_tensor
 
 
 if is_torch_xla_available():
@@ -546,7 +545,7 @@ def prepare_latents(
         return latents, latent_image_ids
 
     # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img.StableDiffusion3Img2ImgPipeline.get_timesteps
-    def get_timesteps(self, num_inference_steps, strength=1.):
+    def get_timesteps(self, num_inference_steps, strength=1.0):
         # get the original timestep using init_timestep
         init_timestep = min(num_inference_steps * strength, num_inference_steps)
 
@@ -583,7 +582,7 @@ def __call__(
         height: Optional[int] = None,
         width: Optional[int] = None,
         eta: float = 1.0,
-        strength: float = 1.,
+        strength: float = 1.0,
         start_timestep: float = 0,
         stop_timestep: float = 0.25,
         num_inference_steps: int = 28,
@@ -843,7 +842,7 @@ def invert(
         source_prompt: str = "",
         source_guidance_scale=0.0,
         num_inversion_steps: int = 28,
-        strength: float = 1.,
+        strength: float = 1.0,
         gamma: float = 0.5,
         height: Optional[int] = None,
         width: Optional[int] = None,
@@ -913,9 +912,7 @@ def invert(
             sigmas,
             mu=mu,
         )
-        timesteps, sigmas, num_inversion_steps = self.get_timesteps(
-            num_inversion_steps, strength
-        )
+        timesteps, sigmas, num_inversion_steps = self.get_timesteps(num_inversion_steps, strength)
 
         # 3. prepare text embeddings
         (
@@ -939,7 +936,7 @@ def invert(
         N = len(sigmas)
 
         # forward ODE loop
-        with self.progress_bar(total=N-1) as progress_bar:
+        with self.progress_bar(total=N - 1) as progress_bar:
             for i in range(N - 1):
                 t_i = torch.tensor(i / (N), dtype=Y_t.dtype, device=device)
                 timestep = torch.tensor(t_i, dtype=Y_t.dtype, device=device).repeat(batch_size)
@@ -969,4 +966,4 @@ def invert(
         self.inverted_latents = Y_t
         self.latent_image_ids = latent_image_ids
 
-        return self.image_latents, Y_t, latent_image_ids
\ No newline at end of file
+        return self.image_latents, Y_t, latent_image_ids

From 8c4d5c1158fa146f6c7a6bcac879dede079701d5 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Wed, 27 Nov 2024 19:00:09 +0200
Subject: [PATCH 28/45] add "copied from..." comments

---
 examples/community/pipeline_flux_rf_inversion.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 77aa6d436ea0..8fb48a304ef0 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -71,7 +71,7 @@
         ```
 """
 
-
+# Copied from diffusers.pipelines.flux.pipeline_flux.calculate_shift
 def calculate_shift(
     image_seq_len,
     base_seq_len: int = 256,
@@ -211,6 +211,7 @@ def __init__(
         )
         self.default_sample_size = 128
 
+    # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline._get_t5_prompt_embeds
     def _get_t5_prompt_embeds(
         self,
         prompt: Union[str, List[str]] = None,

From 3f37448738e7eeb7437e3e88b33f9d3331132e2b Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Wed, 27 Nov 2024 20:54:46 +0200
Subject: [PATCH 29/45] credit authors

---
 examples/community/pipeline_flux_rf_inversion.py | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 8fb48a304ef0..b51287a0f944 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -1,4 +1,6 @@
 # Copyright 2024 Black Forest Labs and The HuggingFace Team. All rights reserved.
+# modeled after RF Inversion: https://rf-inversion.github.io/, authored by Litu Rout, Yujia Chen, Nataniel Ruiz,
+# Constantine Caramanis, Sanjay Shakkottai and Wen-Sheng Chu.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -52,7 +54,10 @@
         >>> import torch
         >>> from diffusers import FluxPipeline
 
-        >>> pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", torch_dtype=torch.bfloat16)
+        >>> pipe = DiffusionPipeline.from_pretrained(
+        ...    "black-forest-labs/FLUX.1-dev",
+        ...    torch_dtype=torch.bfloat16,
+        ...    custom_pipeline="pipeline_flux_rf_inversion")
         >>> pipe.to("cuda")
 
          >>> def download_image(url):
@@ -63,10 +68,14 @@
         >>> img_url = "https://www.aiml.informatik.tu-darmstadt.de/people/mbrack/tennis.jpg"
         >>> image = download_image(img_url)
 
-        >>> _,__,___ = pipe.invert(image=image, num_inversion_steps=28)
+        >>> _,__,___ = pipe.invert(image=image, num_inversion_steps=28, gamma=0.5)
 
         >>> edited_image = pipe(
-        ...     prompt="a portrait of a tiger",
+        ...     prompt="a tomato",
+        ...     start_timestep=0, 
+        ...     stop_timestep=.38,
+        ...     num_inference_steps=28,
+        ...     eta=0.9, 
         ... ).images[0]
         ```
 """

From e06b434a17ba2dde13db1386a9466e0af45712ed Mon Sep 17 00:00:00 2001
From: hlky <hlky@hlky.ac>
Date: Fri, 6 Dec 2024 08:18:57 +0000
Subject: [PATCH 30/45] make style

---
 examples/community/pipeline_flux_rf_inversion.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index b51287a0f944..2e526df92ec7 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -72,14 +72,15 @@
 
         >>> edited_image = pipe(
         ...     prompt="a tomato",
-        ...     start_timestep=0, 
+        ...     start_timestep=0,
         ...     stop_timestep=.38,
         ...     num_inference_steps=28,
-        ...     eta=0.9, 
+        ...     eta=0.9,
         ... ).images[0]
         ```
 """
 
+
 # Copied from diffusers.pipelines.flux.pipeline_flux.calculate_shift
 def calculate_shift(
     image_seq_len,

From e318a6701c4db7a39cf60a30bf04224e80015997 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Fri, 6 Dec 2024 15:17:24 +0200
Subject: [PATCH 31/45] return inversion outputs without self-assigning

---
 examples/community/pipeline_flux_rf_inversion.py | 15 +++++----------
 1 file changed, 5 insertions(+), 10 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 2e526df92ec7..fee9ae4bb80f 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -588,6 +588,9 @@ def interrupt(self):
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
         self,
+        latents: Optional[torch.FloatTensor] = None,
+        image_latents: Optional[torch.FloatTensor] = None,
+        latent_image_ids: Optional[torch.FloatTensor] = None,
         prompt: Union[str, List[str]] = None,
         prompt_2: Optional[Union[str, List[str]]] = None,
         height: Optional[int] = None,
@@ -601,7 +604,6 @@ def __call__(
         guidance_scale: float = 3.5,
         num_images_per_prompt: Optional[int] = 1,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        latents: Optional[torch.FloatTensor] = None,
         prompt_embeds: Optional[torch.FloatTensor] = None,
         pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
         output_type: Optional[str] = "pil",
@@ -735,9 +737,6 @@ def __call__(
 
         # 4. Prepare latent variables
         num_channels_latents = self.transformer.config.in_channels // 4
-        latents = self.inverted_latents
-        latent_image_ids = self.latent_image_ids
-        image_latents = self.image_latents
 
         # 5. Prepare timesteps
         sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
@@ -859,7 +858,6 @@ def invert(
         width: Optional[int] = None,
         timesteps: List[int] = None,
         dtype: Optional[torch.dtype] = None,
-        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
         joint_attention_kwargs: Optional[Dict[str, Any]] = None,
     ):
         r"""
@@ -903,7 +901,6 @@ def invert(
         image_latents, latent_image_ids = self.prepare_latents(
             batch_size, num_channels_latents, height, width, dtype, device, image_latents
         )
-        self.image_latents = image_latents.clone()
 
         # 2. prepare timesteps
         sigmas = np.linspace(1.0, 1 / num_inversion_steps, num_inversion_steps)
@@ -974,7 +971,5 @@ def invert(
                 Y_t = Y_t + u_hat_t_i * (sigmas[i] - sigmas[i + 1])
                 progress_bar.update()
 
-        self.inverted_latents = Y_t
-        self.latent_image_ids = latent_image_ids
-
-        return self.image_latents, Y_t, latent_image_ids
+        # return the inverted latents (start point for the denoising loop), encoded image & latent image ids
+        return Y_t, image_latents, latent_image_ids

From d523e2bba7f887446918cc43ec1a12ef4f5af262 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Fri, 6 Dec 2024 15:33:51 +0200
Subject: [PATCH 32/45] adjust denoising loop to generate regular images if
 inverted latents are not provided

---
 .../community/pipeline_flux_rf_inversion.py   | 103 +++++++++++++++---
 1 file changed, 86 insertions(+), 17 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index fee9ae4bb80f..02b46d36d7ae 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -419,6 +419,9 @@ def check_inputs(
         self,
         prompt,
         prompt_2,
+        inverted_latents,
+        image_latents,
+        latent_image_ids,
         height,
         width,
         start_timestep,
@@ -467,6 +470,10 @@ def check_inputs(
         if max_sequence_length is not None and max_sequence_length > 512:
             raise ValueError(f"`max_sequence_length` cannot be greater than 512 but is {max_sequence_length}")
 
+        if inverted_latents is not None and (image_latents is None or latent_image_ids is None):
+            raise ValueError(
+                "If `inverted_latents` are provided, `image_latents` and `latent_image_ids` also have to be passed. "
+            )
         # check start_timestep and stop_timestep
         if start_timestep < 0 or start_timestep > stop_timestep:
             raise ValueError(f"`start_timestep` should be in [0, stop_timestep] but is {start_timestep}")
@@ -536,7 +543,7 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    def prepare_latents(
+    def prepare_latents_inversion(
         self,
         batch_size,
         num_channels_latents,
@@ -555,6 +562,41 @@ def prepare_latents(
 
         return latents, latent_image_ids
 
+    def prepare_latents(
+        self,
+        batch_size,
+        num_channels_latents,
+        height,
+        width,
+        dtype,
+        device,
+        generator,
+        latents=None,
+    ):
+        # VAE applies 8x compression on images but we must also account for packing which requires
+        # latent height and width to be divisible by 2.
+        height = 2 * (int(height) // (self.vae_scale_factor * 2))
+        width = 2 * (int(width) // (self.vae_scale_factor * 2))
+
+        shape = (batch_size, num_channels_latents, height, width)
+
+        if latents is not None:
+            latent_image_ids = self._prepare_latent_image_ids(batch_size, height // 2, width // 2, device, dtype)
+            return latents.to(device=device, dtype=dtype), latent_image_ids
+
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+        latents = self._pack_latents(latents, batch_size, num_channels_latents, height, width)
+
+        latent_image_ids = self._prepare_latent_image_ids(batch_size, height // 2, width // 2, device, dtype)
+
+        return latents, latent_image_ids
+
     # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img.StableDiffusion3Img2ImgPipeline.get_timesteps
     def get_timesteps(self, num_inference_steps, strength=1.0):
         # get the original timestep using init_timestep
@@ -588,11 +630,11 @@ def interrupt(self):
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
         self,
-        latents: Optional[torch.FloatTensor] = None,
-        image_latents: Optional[torch.FloatTensor] = None,
-        latent_image_ids: Optional[torch.FloatTensor] = None,
         prompt: Union[str, List[str]] = None,
         prompt_2: Optional[Union[str, List[str]]] = None,
+        inverted_latents: Optional[torch.FloatTensor] = None,
+        image_latents: Optional[torch.FloatTensor] = None,
+        latent_image_ids: Optional[torch.FloatTensor] = None,
         height: Optional[int] = None,
         width: Optional[int] = None,
         eta: float = 1.0,
@@ -604,6 +646,7 @@ def __call__(
         guidance_scale: float = 3.5,
         num_images_per_prompt: Optional[int] = 1,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
         prompt_embeds: Optional[torch.FloatTensor] = None,
         pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
         output_type: Optional[str] = "pil",
@@ -693,6 +736,9 @@ def __call__(
         self.check_inputs(
             prompt,
             prompt_2,
+            inverted_latents,
+            image_latents,
+            latent_image_ids,
             height,
             width,
             start_timestep,
@@ -706,6 +752,7 @@ def __call__(
         self._guidance_scale = guidance_scale
         self._joint_attention_kwargs = joint_attention_kwargs
         self._interrupt = False
+        do_rf_inversion = inverted_latents is not None
 
         # 2. Define call parameters
         if prompt is not None and isinstance(prompt, str):
@@ -737,6 +784,19 @@ def __call__(
 
         # 4. Prepare latent variables
         num_channels_latents = self.transformer.config.in_channels // 4
+        if do_rf_inversion:
+            latents = inverted_latents
+        else:
+            latents, latent_image_ids = self.prepare_latents(
+                batch_size * num_images_per_prompt,
+                num_channels_latents,
+                height,
+                width,
+                prompt_embeds.dtype,
+                device,
+                generator,
+                latents,
+            )
 
         # 5. Prepare timesteps
         sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
@@ -769,9 +829,11 @@ def __call__(
         else:
             guidance = None
 
+        if do_rf_inversion:
+            y_0 = image_latents.clone()
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:
-            y_0 = image_latents.clone()
+
             for i, t in enumerate(timesteps):
                 t_i = 1 - t / 1000
                 dt = torch.tensor(1 / (len(timesteps) - 1), device=device)
@@ -782,7 +844,7 @@ def __call__(
                 # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
                 timestep = t.expand(latents.shape[0]).to(latents.dtype)
 
-                v_t = -self.transformer(
+                noise_pred = self.transformer(
                     hidden_states=latents,
                     timestep=timestep / 1000,
                     guidance=guidance,
@@ -794,18 +856,25 @@ def __call__(
                     return_dict=False,
                 )[0]
 
-                v_t_cond = (y_0 - latents) / (1 - t_i)
-                eta_t = eta if start_timestep <= i < stop_timestep else 0.0
-                if start_timestep <= i < stop_timestep:
-                    # controlled vector field
-                    v_hat_t = v_t + eta * (v_t_cond - v_t)
+                if do_rf_inversion:
+                    v_t = -noise_pred
 
-                else:
-                    v_hat_t = v_t
-                # SDE Eq: 17
+                    v_t_cond = (y_0 - latents) / (1 - t_i)
+                    eta_t = eta if start_timestep <= i < stop_timestep else 0.0
+                    if start_timestep <= i < stop_timestep:
+                        # controlled vector field
+                        v_hat_t = v_t + eta * (v_t_cond - v_t)
 
-                latents_dtype = latents.dtype
-                latents = latents + v_hat_t * (sigmas[i] - sigmas[i + 1])
+                    else:
+                        v_hat_t = v_t
+                    # SDE Eq: 17
+
+                    latents_dtype = latents.dtype
+                    latents = latents + v_hat_t * (sigmas[i] - sigmas[i + 1])
+                else:
+                    # compute the previous noisy sample x_t -> x_t-1
+                    latents_dtype = latents.dtype
+                    latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
 
                 if latents.dtype != latents_dtype:
                     if torch.backends.mps.is_available():
@@ -898,7 +967,7 @@ def invert(
 
         # 1. prepare image
         image_latents, _ = self.encode_image(image, height=height, width=width, dtype=dtype)
-        image_latents, latent_image_ids = self.prepare_latents(
+        image_latents, latent_image_ids = self.prepare_latents_inversion(
             batch_size, num_channels_latents, height, width, dtype, device, image_latents
         )
 

From 6d98da3bb3cb5a40a55e1ecc572b96ce9df7d2bc Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Fri, 6 Dec 2024 15:36:30 +0200
Subject: [PATCH 33/45] adjust denoising loop to generate regular images if
 inverted latents are not provided

---
 examples/community/pipeline_flux_rf_inversion.py | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 02b46d36d7ae..c7498f91cf59 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -799,7 +799,7 @@ def __call__(
             )
 
         # 5. Prepare timesteps
-        sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
+        sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else sigmas
         image_seq_len = (int(height) // self.vae_scale_factor // 2) * (int(width) // self.vae_scale_factor // 2)
         mu = calculate_shift(
             image_seq_len,
@@ -816,9 +816,10 @@ def __call__(
             sigmas,
             mu=mu,
         )
-        start_timestep = int(start_timestep * num_inference_steps)
-        stop_timestep = min(int(stop_timestep * num_inference_steps), num_inference_steps)
-        timesteps, sigmas, num_inference_steps = self.get_timesteps(num_inference_steps, strength)
+        if do_rf_inversion:
+            start_timestep = int(start_timestep * num_inference_steps)
+            stop_timestep = min(int(stop_timestep * num_inference_steps), num_inference_steps)
+            timesteps, sigmas, num_inference_steps = self.get_timesteps(num_inference_steps, strength)
         num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
         self._num_timesteps = len(timesteps)
 
@@ -833,10 +834,10 @@ def __call__(
             y_0 = image_latents.clone()
         # 6. Denoising loop
         with self.progress_bar(total=num_inference_steps) as progress_bar:
-
             for i, t in enumerate(timesteps):
-                t_i = 1 - t / 1000
-                dt = torch.tensor(1 / (len(timesteps) - 1), device=device)
+                if do_rf_inversion:
+                    t_i = 1 - t / 1000
+                    dt = torch.tensor(1 / (len(timesteps) - 1), device=device)
 
                 if self.interrupt:
                     continue

From 37b70a716562a0eb0a60fe130a2721777593d739 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Fri, 6 Dec 2024 15:52:11 +0200
Subject: [PATCH 34/45] fix import

---
 examples/community/pipeline_flux_rf_inversion.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index c7498f91cf59..9b46dfb24aba 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -36,6 +36,7 @@
     scale_lora_layers,
     unscale_lora_layers,
 )
+from diffusers.utils.torch_utils import randn_tensor
 
 
 if is_torch_xla_available():
@@ -642,6 +643,7 @@ def __call__(
         start_timestep: float = 0,
         stop_timestep: float = 0.25,
         num_inference_steps: int = 28,
+        sigmas: Optional[List[float]] = None,
         timesteps: List[int] = None,
         guidance_scale: float = 3.5,
         num_images_per_prompt: Optional[int] = 1,

From c4ce7401857640d59330b78188e28ab5297b54e7 Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Fri, 6 Dec 2024 17:30:33 +0200
Subject: [PATCH 35/45] comment

---
 examples/community/pipeline_flux_rf_inversion.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 9b46dfb24aba..0c8b40ebf276 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -870,8 +870,8 @@ def __call__(
 
                     else:
                         v_hat_t = v_t
-                    # SDE Eq: 17
 
+                    # SDE Eq: 17 from https://arxiv.org/pdf/2410.10792
                     latents_dtype = latents.dtype
                     latents = latents + v_hat_t * (sigmas[i] - sigmas[i + 1])
                 else:

From a61c93fed3846d7f85fe7afb86527707d610333b Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Fri, 6 Dec 2024 17:31:25 +0200
Subject: [PATCH 36/45] remove redundant line

---
 examples/community/pipeline_flux_rf_inversion.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 0c8b40ebf276..c153913031bc 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -859,6 +859,7 @@ def __call__(
                     return_dict=False,
                 )[0]
 
+                latents_dtype = latents.dtype
                 if do_rf_inversion:
                     v_t = -noise_pred
 
@@ -872,11 +873,9 @@ def __call__(
                         v_hat_t = v_t
 
                     # SDE Eq: 17 from https://arxiv.org/pdf/2410.10792
-                    latents_dtype = latents.dtype
                     latents = latents + v_hat_t * (sigmas[i] - sigmas[i + 1])
                 else:
                     # compute the previous noisy sample x_t -> x_t-1
-                    latents_dtype = latents.dtype
                     latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
 
                 if latents.dtype != latents_dtype:

From 9116a056ec12665f849359f01892dc04e142c91f Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Fri, 6 Dec 2024 18:01:02 +0200
Subject: [PATCH 37/45] modify comment on ti

---
 examples/community/pipeline_flux_rf_inversion.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index c153913031bc..c1fed43150fe 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -834,10 +834,11 @@ def __call__(
 
         if do_rf_inversion:
             y_0 = image_latents.clone()
-        # 6. Denoising loop
+        # 6. Denoising loop / Controlled Reverse ODE, Algorithm 2 from: https://arxiv.org/pdf/2410.10792
         with self.progress_bar(total=num_inference_steps) as progress_bar:
             for i, t in enumerate(timesteps):
                 if do_rf_inversion:
+                    # ti (current timestep) as annotated in algorithm 2 - i/num_inference_steps.
                     t_i = 1 - t / 1000
                     dt = torch.tensor(1 / (len(timesteps) - 1), device=device)
 
@@ -862,7 +863,6 @@ def __call__(
                 latents_dtype = latents.dtype
                 if do_rf_inversion:
                     v_t = -noise_pred
-
                     v_t_cond = (y_0 - latents) / (1 - t_i)
                     eta_t = eta if start_timestep <= i < stop_timestep else 0.0
                     if start_timestep <= i < stop_timestep:

From e87d45a865115ab4d8d305c44518c6d5d3a43a73 Mon Sep 17 00:00:00 2001
From: Linoy Tsaban <57615435+linoytsaban@users.noreply.github.com>
Date: Mon, 9 Dec 2024 18:42:10 +0200
Subject: [PATCH 38/45] Update examples/community/pipeline_flux_rf_inversion.py

Co-authored-by: hlky <hlky@hlky.ac>
---
 examples/community/pipeline_flux_rf_inversion.py | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index c1fed43150fe..4d27d79287e7 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -69,10 +69,13 @@
         >>> img_url = "https://www.aiml.informatik.tu-darmstadt.de/people/mbrack/tennis.jpg"
         >>> image = download_image(img_url)
 
-        >>> _,__,___ = pipe.invert(image=image, num_inversion_steps=28, gamma=0.5)
+        >>> inverted_latents, image_latents, latent_image_ids = pipe.invert(image=image, num_inversion_steps=28, gamma=0.5)
 
         >>> edited_image = pipe(
         ...     prompt="a tomato",
+        ...     inverted_latents=inverted_latents,
+        ...     image_latents=image_latents,
+        ...     latent_image_ids=latent_image_ids,
         ...     start_timestep=0,
         ...     stop_timestep=.38,
         ...     num_inference_steps=28,

From f8aeb1681cfde0019cc146b3687e3a532daf2f0e Mon Sep 17 00:00:00 2001
From: Linoy Tsaban <57615435+linoytsaban@users.noreply.github.com>
Date: Mon, 9 Dec 2024 18:42:45 +0200
Subject: [PATCH 39/45] Update examples/community/pipeline_flux_rf_inversion.py

Co-authored-by: hlky <hlky@hlky.ac>
---
 examples/community/pipeline_flux_rf_inversion.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 4d27d79287e7..003722c3b90e 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -80,6 +80,9 @@
         ...     stop_timestep=.38,
         ...     num_inference_steps=28,
         ...     eta=0.9,
+        ...     stop_timestep=.38,
+        ...     num_inference_steps=28,
+        ...     eta=0.9,
         ... ).images[0]
         ```
 """

From f290b242b8b9a950eccca3d75c2cc4bcff91befb Mon Sep 17 00:00:00 2001
From: Linoy Tsaban <57615435+linoytsaban@users.noreply.github.com>
Date: Mon, 9 Dec 2024 18:43:00 +0200
Subject: [PATCH 40/45] Update examples/community/pipeline_flux_rf_inversion.py

Co-authored-by: hlky <hlky@hlky.ac>
---
 examples/community/pipeline_flux_rf_inversion.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 003722c3b90e..ab9d6de00559 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -322,6 +322,7 @@ def _get_clip_prompt_embeds(
 
         return prompt_embeds
 
+    # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline.encode_prompt
     def encode_prompt(
         self,
         prompt: Union[str, List[str]],

From e42dd6bcf322147f3b5cd3a00ae5ec46e88153e3 Mon Sep 17 00:00:00 2001
From: Linoy Tsaban <57615435+linoytsaban@users.noreply.github.com>
Date: Mon, 9 Dec 2024 18:43:29 +0200
Subject: [PATCH 41/45] Update examples/community/pipeline_flux_rf_inversion.py

Co-authored-by: hlky <hlky@hlky.ac>
---
 examples/community/pipeline_flux_rf_inversion.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index ab9d6de00559..90a85e1f6ba6 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -278,6 +278,7 @@ def _get_t5_prompt_embeds(
 
         return prompt_embeds
 
+    # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline._get_clip_prompt_embeds
     def _get_clip_prompt_embeds(
         self,
         prompt: Union[str, List[str]],

From 0036ae2a3c13f86747eab1fb91e452110f57e121 Mon Sep 17 00:00:00 2001
From: Linoy Tsaban <57615435+linoytsaban@users.noreply.github.com>
Date: Mon, 9 Dec 2024 18:44:27 +0200
Subject: [PATCH 42/45] Update examples/community/pipeline_flux_rf_inversion.py

Co-authored-by: hlky <hlky@hlky.ac>
---
 examples/community/pipeline_flux_rf_inversion.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 90a85e1f6ba6..38bb08e4c452 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -279,6 +279,7 @@ def _get_t5_prompt_embeds(
         return prompt_embeds
 
     # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline._get_clip_prompt_embeds
+    def _get_clip_prompt_embeds(
     def _get_clip_prompt_embeds(
         self,
         prompt: Union[str, List[str]],

From 074cf5adfe4ab331a4b882c2a2824defc23f96df Mon Sep 17 00:00:00 2001
From: Linoy Tsaban <57615435+linoytsaban@users.noreply.github.com>
Date: Mon, 9 Dec 2024 18:45:39 +0200
Subject: [PATCH 43/45] Update examples/community/pipeline_flux_rf_inversion.py

Co-authored-by: hlky <hlky@hlky.ac>
---
 examples/community/pipeline_flux_rf_inversion.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 38bb08e4c452..ae1a44e3fdf5 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -572,6 +572,7 @@ def prepare_latents_inversion(
 
         return latents, latent_image_ids
 
+    # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline.prepare_latents
     def prepare_latents(
         self,
         batch_size,

From 5f1407dc8e49b3d1a450ebe1e76ea60ea9b4d023 Mon Sep 17 00:00:00 2001
From: Linoy Tsaban <57615435+linoytsaban@users.noreply.github.com>
Date: Mon, 9 Dec 2024 18:46:18 +0200
Subject: [PATCH 44/45] Update examples/community/pipeline_flux_rf_inversion.py

Co-authored-by: hlky <hlky@hlky.ac>
---
 examples/community/pipeline_flux_rf_inversion.py | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index ae1a44e3fdf5..0a208d4b4b28 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -678,6 +678,12 @@ def __call__(
             prompt_2 (`str` or `List[str]`, *optional*):
                 The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
                 will be used instead
+            inverted_latents (`torch.Tensor`, *optional*):
+                The inverted latents from `pipe.invert`.
+            image_latents (`torch.Tensor`, *optional*):
+                The image latents from `pipe.invert`.
+            latent_image_ids (`torch.Tensor`, *optional*):
+                The latent image ids from `pipe.invert`.
             height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                 The height in pixels of the generated image. This is set to 1024 by default for the best results.
             width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):

From 1cb1ce62871406d4a7ca36d5ce1eca182129257b Mon Sep 17 00:00:00 2001
From: linoytsaban <linoy@huggingface.co>
Date: Mon, 9 Dec 2024 18:52:16 +0200
Subject: [PATCH 45/45] fix syntax error

---
 examples/community/pipeline_flux_rf_inversion.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/examples/community/pipeline_flux_rf_inversion.py b/examples/community/pipeline_flux_rf_inversion.py
index 0a208d4b4b28..7f5f1b02695e 100644
--- a/examples/community/pipeline_flux_rf_inversion.py
+++ b/examples/community/pipeline_flux_rf_inversion.py
@@ -279,7 +279,6 @@ def _get_t5_prompt_embeds(
         return prompt_embeds
 
     # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline._get_clip_prompt_embeds
-    def _get_clip_prompt_embeds(
     def _get_clip_prompt_embeds(
         self,
         prompt: Union[str, List[str]],
@@ -723,7 +722,7 @@ def __call__(
                 The output format of the generate image. Choose between
                 [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
             return_dict (`bool`, *optional*, defaults to `True`):
-                Whether or not to return a [`~pipelines.flux.FluxPipelineOutput`] instead of a plain tuple.
+                Whether to return a [`~pipelines.flux.FluxPipelineOutput`] instead of a plain tuple.
             joint_attention_kwargs (`dict`, *optional*):
                 A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
                 `self.processor` in