Can't generate higher resolutions after updating as of 05-29-2023

[ArchAngelAries]

For some reason I'm no longer able to generate anything higher than 768 x 768... On my last install I could generate up to 1024 x 1536. I'm on a 7900 XT. Any help would be greatly appreciated.

my webui-user.bat settings:

set PYTHON=
set GIT=
set VENV_DIR=
set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:512
set COMMANDLINE_ARGS=--medvram --no-half --no-half-vae --opt-split-attention --opt-sub-quad-attention --disable-nan-check --precision full --autolaunch
set SAFETENSORS_FAST_GPU=1

call webui.bat

[lshqqytiger]

It might be related: AUTOMATIC1111#10763

[songib]

@lshqqytiger
I think this is still a problem atm since I can't generate any images after out of memory and --lowvram didn't help after you tried to generate bigger images and I needed to restart the WebUI afterward. And I still can't use the Adetailer either.

Is there any solution?

[Siralob]

@songib
Adetailer works with the latest commit 69c8fae with --medvram.
I noticed Adetailer didn't work correctly even it detected face and also in-painted in the live preview. But the output was the same as input blurry face.
So, I did update Adetailer to f18e00b, also added --no-half in webui-user.bat. It works now. (deleted no-half-vae)

Thank you @Miraihi, I don't need to put --no-half-vae in command-line arguments.

I'm also wondering why vladmandic/automatic can use adetailer without these arguments in setting tab. I unchecked box related FP32 computing... idk why, webui force to use them internally?

[Miraihi]

Inpainting doesn't work without --no-half for most people, so it's to be expected that the case is the same for Adetailer.

[Miraihi]

After the latest commit I can't use ESRGAN-based upscalers at all. Throws me the "out of memory" error. Also the black squares of SCUnet return.

[Miraihi]

Update: Upscaling and img2img works again after fiddling with the new "Optimizations" settings.
Delete --opt-split-attention --opt-sub-quad-attention from your webui-user.bat and try something like this.

[ArchAngelAries]

Update: Upscaling and img2img works again after fiddling with the new "Optimizations" settings. Delete --opt-split-attention --opt-sub-quad-attention from your webui-user.bat and try something like this.

Tried this and just got "RuntimeError: mat1 and mat2 must have the same dtype"

[howDoILearnPython]

then can we, the AMD card user, use the Doggettx optimization to temporarily solve the problem??

[howDoILearnPython]

tried. i got similar VRAM usage when i select sub-quadratic in the optimizations page inside setting.