[bug]: blotchy images after converting ckpt/safetensors to diffusers

[jonathanljs]

Is there an existing issue for this?

• [X] I have searched the existing issues

OS

Windows

GPU

cuda

VRAM

8GB

What happened?

All diffuser models come out blotchy to various levels (original or converted) - see attached screenshot. Doesn't happen with Safetensors / CKPT models.

Screenshots

Additional context

I'm on an older GPU (1070) and there are some models it can't cope with at all. Could this be the source of the problem?

Contact Details

discordname jonathanljs#0290

[cwallen]

I've seen this as well though not nearly as bad as your screenshot. I think it only happens on some samplers.

[lstein]

Could you walk through the exact sequence of steps you took that leads to this result? So far I haven’t been able to replicate it?

[jonathanljs]

It was quite simple: I converted some ckpt/safetensor models to diffusers using the command line, and when I created a new image, that's what I got when doing txt2img or img2img. Also with the inbuilt diffuser models. On 2.3.0. Now I'm on 2.3.1 and diffusers still don't work, I just get semi-blank fog on txt2img and foggy versions of the original on img2img. Some ckpt/safetensors also don't work on my system, often seems to be the fp32 ones. I'm thinking it might be because I'm using a GTX1070 and it just can't handle it..? (I have xformers running.) don't know if it's connected, but Python 3.10 doesn't work in any form either, only 3.9. This is a Windows 11 machine.

[github-actions[bot]]

There has been no activity in this issue for 14 days. If this issue is still being experienced, please reply with an updated confirmation that the issue is still being experienced with the latest release.