Closed Enferlain closed 1 year ago
Three possible issues:
Try running without animeVae and make sure to select NovelAI. Otherwise, there should be no difference as the models have remained unchanged. If there is an optimizer that lowers quality, check out issues in AUTOMATIC1111/stable-diffusion-webui
Tried running with just novelai enabled but still looks bad. Probably something with webui itself I guess. Checked open issues and this is the only similar thing I found
Unless something got enabled/disabled that now requires manual input to change, I have no clue and would probably struggle to find out what exactly.
Looked a bit more, and these 2 might be related. Not sure if they got fixed since
https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/3343 https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/3145#issue-1414833114
How is it if you disable highres fix?
Fixed by AUTOMATIC
How/where was it fixed?
How/where was it fixed?
Not by me. The problem was with the upstream WebUI
Yes, obviously, but you have no idea what broke or when/where it was fixed? None of the other issues on the webui know either. You're the first to say it was fixed.
Because it no longer spits out blurry images (Subjective and based on my own testing)
If you continue to encounter the issue, feel free to make me reopen it
Haven't used the colab in like 5 days and seems like there have been some updates to it. The thing I noticed right away is that default quality seems to be worse. Prompts and img2img seem to have similar results but the end product is lower in quality, like blurry/bad faces, hair, hands are almost always bad now, and other details that used to be fine.
Is there a way to test and compare to like October 20 and before vs oct 25? I don't think I'm tripping but I'd like to confirm.
Like here I'm using pretty much the same prompts but the quality is way different