FAQ & Troubleshooting
My output imgs look completely different to each other, like it's doing pure txt2img and not img2img
- Go to your webui settings --> look for this setting, and make sure it's not checked:
- The social media platforms are compressing the uploaded videos, so if you upload one (at low res), it will be compressed even further resulting in the unrecognizable goo. To prevent it, please upscale the videos at least 'x2' in the Output tab before posting anywhere except for Discord.
While we try our best to keep up with the latest CN updates, Deforum is not responsible for the ControlNet extension. Please, report the bugs here instead https://github.com/Mikubill/sd-webui-controlnet/issues !
- If you got the
AttributeError: 'NoneType' object has no attribute 'get', it means that one of the 3D model files is either missing, or downloaded only partially. The fix is to manually download the models again, and putting both of them in the/models/Deforumfolder.
- Download AdaBins_nyu.pt from here
- Download dpt_large-midas-2f21e586.pt from here
- You have a missing comma (,) after one of your prompts. Make sure all of the prompts EXCEPT for the last one have a trailing comma after them.
- Should look like this:
{"0": "Cool prompt at frame 0","60": "Cool prompt at frame 60","80": "Cool prompt at frame 80","100": "Cool prompt at frame 100"}
and NOT like this (note the missing comma after the second prompt):
{"0": "Cool prompt at frame 0","60": "Cool prompt at frame 60" "80": "Cool prompt at frame 80","100": "Cool prompt at frame 100"}
Most likely, it happened because it didn't contain the full crash report. Good bugreports bot only have the error message and the traceback, but also all the info about your environment and all the other extensions, extra networks, etc. you're using. If the devs won't know all the information, it may lead them to chasing a ghost and simply frustrate everybody. Remember, «the software is provided "as is", without warranty of any kind» and the devs are all volunteers. So if you want to have your problem fixed, please at least provide it an unambiguous description.
TL;DR: your log should start like this
PS C:\Users\kabachuha\stable-diffusion-webui> .\webui-user.bat
venv "C:\Users\kabachuha\stable-diffusion-webui\venv\Scripts\Python.exe"
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: v1.3.1
Commit hash: b6af0a3809ea869fb180633f9affcae4b199ffcf
Installing requirements
Launching Web UI with arguments: --medvram --no-half-vae
...
...
...
Like this:
- Most likely, is was caused by setting the 'Seed behaviour' to 'Fixed'. If the seed is the same, denoising doesn't happen and the image gets covered with noise over time. Please, don't use this option unless you know what you're doing. For beginners, it's better to use any other seed behavior, the simplest is 'Iter'.
No, sorry.
Like this:
Or:
- There are
twomain ways of fixing it: either lowering thestrength scheduleor heightening thenoise multiplier schedule... or using ControlNet.
The most luxurious and the most efficient is ControlNet in Loopback-mode. This way it feed the each frame as a ControlNet base for the next frame, preserving the details in a subtle way.
As for the other two ways
-
These values are a good rule of thumb, but be cautious when mixing them together
-
Strength schedule for arbitrary number of frames can look like this:
0:(0.25 * cos((72 / 60 * 3.141 * (t + 0) / 30))**13 + 0.7)
Example of strength drop (set up manually by mechakull on Discord):
0: (0.55), 20: (0.6), 30: (0.6), 41: (0.3), 60: (0.6), 70: (0.6), 81: (0.3), 100: (0.6), 110: (0.6), 121: (0.3), 140: (0.6), 150: (0.6), 161: (0.35), 180: (0.6), 200: (0.6), 201: (0.3), 220: (0.6), 230: (0.6), 241: (0.3), 260: (0.6), 270: (0.6), 281: (0.35), 300: (0.6), 310: (0.6), 321: (0.35), 340: (0.6), 350: (0.6), 361: (0.3), 380: (0.6), 401:(0.3), 420: (0.6), 430: (0.6), 441: (0.35), 460: (0.6, 470: (0.6), 481: (0.3), 500: (0.6), 510: (0.6), 521: (0.3), 540: (0.6), 550: (0.6), 561: (0.3), 580: (0.6), 600:(0.3)
- Noise multiplier:
0: (1.07-(abs(sin(2*3.14*t/5+0))*0.02))
Example of increased noise multiplier:
Like this:
- Lower Anti-Blur's
amount_scheduleornoise multiplier scheduleif you have it enabled.
Deforum crashes with KeyError after I press 'Generate'. It seems it is related to the keyframes you added recently.
File "/home/kabachuha/kml/kek/stable-diffusion-webui/extensions/deforum/scripts/deforum_helpers/animation_key_frames.py", line 9, in __init__
self.angle_series = get_inbetweens(parse_key_frames(anim_args.angle), anim_args.max_frames)
File "/home/kabachuha/kml/kek/stable-diffusion-webui/extensions/deforum/scripts/deforum_helpers/animation_key_frames.py", line 79, in get_inbetweens
key_frame_series[0] = key_frame_series[key_frame_series.first_valid_index()]
File "/home/kabachuha/.local/lib/python3.8/site-packages/pandas/core/series.py", line 982, in __getitem__
return self._get_value(key)
File "/home/kabachuha/.local/lib/python3.8/site-packages/pandas/core/series.py", line 1092, in _get_value
loc = self.index.get_loc(label)
File "/home/kabachuha/.local/lib/python3.8/site-packages/pandas/core/indexes/range.py", line 396, in get_loc
raise KeyError(key)
KeyError: None
- Go to webui's core folder and remove ui-config.json. This action will clear the cache and it often solves this issue (especially if no one else seems to have it, as each other's cache forms in a different way)
There are two main ways to do so:
- Use the 'Guided images' mode in the extension.
- Interpolate the pics using a frame interpolation engine in the 'Output' tab.
The 'Guided images' mode slowly blends in the target images into the usual Deforum animation process using a complicated math formula, before the diffusion starts to remake the images. This way the images 'guide' the AI to morph the frames being made towards the intended shapes.
Frames used:
There are two main interpolation engines: RIFE and FILM. FILM is slower and demands more resources, but it's worth it.
Insert the pics or a video into the output tab, select the needed settings and interpolate the pics of your choice, even not AI-generated!
Frames used:
Naturally, interpolate your video before upscaling or else it will take a ton of time.
As of now, there are NO official Deforum-generation Discord bots. That being said, sorry, but you'll have to generate the images on your own PC or in a cloud.
... Unless you donate us enough money to host one, wink-wink
This is a known issue with no known fix as of January 19th 2023. Please wait for a fix. Issue can be tracked Fixed with an update of the core auto. If you have the issue, upgrade your webui.
- There was a button, but it turned out what it applied only to the first frame of the animation, thus it used to create only an illusion of usefulness. We decided to remove it not to confuse the users and to free the UI space.
How do I protect my animations from duplicating the content then, such as making multiple heads or even splitting the horizon?
Please, render it on lesser resolution. (the greater dim should be not much more than 512 for SD 1.X. 728 for 2.X) You can always upscale your videos using Deforum's internal upscalers in the Output tab.