support diffusers
akk-123 opened this issue · 6 comments
The current code is a bit bloated, can you provide more concise diffusers code?
Thanks for your attention! In the near future, we will provide diffusers-based code.
The current code is a bit bloated, can you provide more concise diffusers code?
Thanks for your waiting, we now have supported diffusers.
@YangLing0818 I test it, but I get a black img, I found latents here is all NaN
I test with model runwayml/stable-diffusion-v1-5 and use gpt-4
Thank you for your feedback. If you get black image, it may because of the NSFW filter. The prompt you have given or generated image might be closer to one of the not allowed content. In such cases you can get black image as output . You can use different prompt and try different random seed to test if it is becasue of the NSFW filter. And it could also be becasue of your device , which may not support half-precision, try to use full-precision instead.
Not because it’s NSFW, I set safe_checker is None, and I sad that latents is also NAN, I try fp32, It's same
Thank you for your feedback ! I have tested the code in RPG.py on different devices including A100-80G, A100-40G and RTX4090. For me, model runwayml/stable-diffusion-v1-5 in full-precision and half-precision works well with A100, and RTX4090 works well with full-precision. Would you mind running diffusers-cli env and pasting the output from the command? Also, please let us know the GPU and the code you are using . Thank you! You can also refer to the issue which has been handled here huggingface/diffusers#2153