easydiffusion/easydiffusion

render issue

Sirmadsen opened this issue · 5 comments

I am getting a render issue that I cannot seem to fix. On other Guis it's usually on my end a checkpoint or built in keywords (Fooocus) that can be changed to fix it.

Here, I'm getting this issue no matter what I do. I have tried:

  • Multiple different checkpoints (SD and SDXL)
  • Different variation scales
  • With/Without loras
  • Different resolutions
  • Different clip skip
  • Different prompts
  • With and without negative prompt
  • Different samplers
  • Different GPU memory usage settings
  • Different Browsers

I only get the dreaded color image no matter what I do. It worked on the first 10 images then I get this issue without making any significant changes. I even tried installing on another PC and got the issue directly.
Any tips?

2024-06-02 20_17_46-Easy Diffusion

Hi @Sirmadsen I know you've tried different combinations, but can you please try this exact combination:

Model Name: sd-v1-5 (the one that's included with Easy Diffusion)
Steps: 25
Sampler: Euler Ancestral
Guidance Scale: 7.5
Prompt: Astronaut riding a horse
Size: 512 x 512

No LoRAs, no embeddings, no clip skip. Also, no UI plugins.

Thanks

Well, I have been testing all kind of things, default settings included. I decided to completely reinstall python and other dependencies as well... Now it works. It still starts out as the image above but slowly renders into the prompt and generated the images in the last 20-30% of the steps. Maybe I screwed something up in the settings before. I'm guessing the renderer works different than Fooocus that I'm used to work in where you see the build up of the images from scratch instead of the color noise slowly building into the image.

So, it works now as intended.

I do have an unrelated question though. Where do I put upscale models? I've tried to put them in the realesgran folder with no luck.

It still starts out as the image above but slowly renders into the prompt and generated the images in the last 20-30% of the steps

instead of the color noise slowly building into the image

That's dependent on the particular sampler. Different samplers start with different initial images.

Where do I put upscale models

Sorry, ED doesn't support custom upscaling models right now

Thank you very much for the help and taking time to answer my questions. It's a nice gui and I very much like the design. Keep up the good work.

Sorry, ED doesn't support custom upscaling models right now

Aww, hopefully one day xD