Won-Kyu Park
Won-Kyu Park
> Are AI Toolkit loras supported at all? There is _some_ effect as the generated images are different with and without them, but not the expected effect, and there are...
> And [this lora](https://civitai.com/models/424599?modelVersionId=473067) seems to always produce NaN's in Unet when the patch is applied but works fine without it. works just fine for me with no issue. ~~what...
> I think it's the new setting `LoRA without backup weights` that affects it. If it's off I can generate without problems, if it's on some loras break like that....
the last commit have some issue. will try to fix soon.
to use `lora_without_backup_weights` correctly, we can't use `torch.autocast()` at all. so `devices.autocast()` has been fixed to replace `torch.autocast()` in this case.
https://github.com/lllyasviel/stable-diffusion-webui-forge/blob/main/webui.bat#L4C5-L4C28 you can add `webui.settings.bat`
to reduce ui loading, mode A loads it's model list by default, all other lists of mode B,C,... and Base model will be lazy loaded after ui starting completed. so,...
`[from:to:when]` prompt syntax could be used for it. for example `[a cat:a dog:0.5] on a beach` see also https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features#prompt-editing
1. do not install or run as `root` user. login as normal user first. 2. do not run `python launch.py ...` directly, use `./webui.sh` shell script. (or `bash webui.sh`) 3....
> I was not able to replicate the issue possible reproduction condition added.