Not working in Automatic 1111
Hey, thank you for the updated version for SD 2.1
Sadly I wasn't able to get it to work yet. Maybe I did something wrong, but I don't know what it could be.
I only downloaded the ckpt-file, renamed it to Avatar21.ckpt and placed it in the models/stablediffusion directory. Then I copied the 2.1 yaml file and gave it the same name.
When I prompt "photorealistic headshot of Will Smith, avatar style" (at both 512 and 768 size) I only get a monochrome, gray image
The old version for 1.5 worked with no issues.
If you have any suggestion what I could try to make it work, I'd very very grateful!
Hello, have you tried cloning the entire repository? I've never used it with just the ckpt file, I always call the entire repository
MODEL_NAME = "Jersonm89/SD2.1_Avatar"
I can of course download all the files, i.e. cloning - but what do I do with it? Where do I put it and what else is needed?
If I have to use it with the original Stable Diffusion, unfortunately I never got it to work. The automatic 1111 webui is the only one that works for me. And for that the ckpt file is enough...
Hey, me again! :)
After a loooot of testing, I found out a few things:
- The monochrome images as mentioned above come from the wrong yaml file.
I used the yaml for ckpt-models that were trained in 768x768. But this model was actually 512x512. So I needed "v2-inference.yaml" from https://github.com/Stability-AI/stablediffusion/tree/main/configs/stable-diffusion instead of "v2-inference-v.yaml".
If someone has the same problem as me, that's the solution.
But I still couldn't create avatar images, because...
- the avatar style was missing from the model.ckpt in your root folder.
Not sure why that is, but it could be that for the way you use it the ckpt file is not used at all.
I was able to convert your repository to a new ckpt file with the dreambooth extension of automatic 111, and indeed this one also included the avatar style :D
kind regards!
it not work