Feedback on Test T2I Cases Gallery

[zdxpan]

here my test t2i cases gallery on figma:

Dear Team,

Here is the link to my test Text-to-Image (T2I) cases gallery on Figma:Lumina image 2.0 VS flux.dev

Overall, the results look promising. However, upon closer inspection, it appears that there is a lack of training in some areas. I am curious about the reasons behind this and would appreciate any insights on how we can improve the model.

Thank you for your attention to this matter.

Best regards,
Alexpan

[gaopengpjlab]

Your feedback is very helpful in identifying the shortcomings of Lumina 2 and in guiding the development of next version.

We share the same observations as you. We hypothesize the following reasons :

1. The 2B parameter is unable to capture detail visual appearance.
2. The training dataset, especially high-quality images, is too small to cover the various aspects of image details.
3. Inadequate training due to the limitation of GPU resources.

We will further investigate these issues in the future and share our insights with community in the future.

[gaopengpjlab]

Can you share your prompts for image generation?

[zdxpan]

sure, here my prompts:

prompts.txt