Open rytt0001 opened 1 year ago
gradio edit image component is inefficient and buggy it uses 5 canvases instead of 1 the input resolution 2160 x 3840 is huge sending 5 times the data to the backend for processing then sending them back to the inpaint view gradio is planning to rewrite the image edit component put there is no milestone for this I am thinking to bypass it and copy the buffers directly without the need to roundtrip the backend
Ya, I'm seeing the same thing, and it essentially causes my browser to stop responding entirely.
Why does this work fine though in upstream, but not in this fork?
after recent commit same issue with text2img , the generate button stops responding when clicked and doesn't launch a new generation. first few generation works fine,( guess it has something to do with queue. )
I've had to move back to automatic1111, but I'm on an older commit (22bcc7be1
Mar 29) for stability reasons.
the info provided is so little to reproduce the issue maybe it is a racing problem you can disable the livepreview from settings or use Aprox NN in preview mode @ghostsquad the commit that works for you has a different gradio version
@anapnoe correct. I have both automatic1111 installed at 22bcc7be1
as well as this fork installed.
any update on this?
Is there an existing issue for this?
What happened?
the generate button of the img2img tab stop responding when clicked and doesn't launch a new generation. It does It in Inpaint mode but not in base Img2Img mode.
Webui is running on linux Manjaro, while accessed on Windows 10.
Steps to reproduce the problem
Go to Img2Img tab then select Inpaint
send an image to the inpaint mode and then draw a mask
to produce the bug you need to write some text in the prompt box and immediately click on the generate button, while the token counter is refreshing
then if done right, the generate will become the interrupt/skip buttons for a few seconds before returning to generate, without launching a new generation
What should have happened?
launched a new generation
Commit where the problem happens
079a5e4d8c8a3a4de43f75a7f64a8f363d02a8c1
What platforms do you use to access the UI ?
Windows
What browsers do you use to access the UI ?
Brave
Command Line Arguments
List of extensions
https://github.com/DominikDoom/a1111-sd-webui-tagcomplete https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111.git https://github.com/ljleb/prompt-fusion-extension.git https://github.com/thomasasfk/sd-webui-aspect-ratio-helper.git https://github.com/Mikubill/sd-webui-controlnet https://github.com/hnmr293/sd-webui-cutoff https://github.com/hako-mikan/sd-webui-regional-prompter.git https://github.com/yfszzx/stable-diffusion-webui-images-browser https://github.com/arenatemp/stable-diffusion-webui-model-toolkit https://github.com/ilian6806/stable-diffusion-webui-state.git https://github.com/Coyote-A/ultimate-upscale-for-automatic1111.git
Console logs
Additional information
It seems the size of the input image affect the chance to produce the bug, when the bug happened I was using a 2160 x 3840 image. same image resized to 1440 x 2560 and the bug doesn't happen for the 10 tries that I did. Using the computer that host the webui to access it (localhost:7860) do not produce the bug.