AUTOMATIC1111 / stable-diffusion-webui

Stable Diffusion web UI
GNU Affero General Public License v3.0
140.83k stars 26.64k forks source link

[Feature Request]: unblend images into diferent images #7805

Open asecino1999 opened 1 year ago

asecino1999 commented 1 year ago

Is there an existing issue for this?

What would your feature do ?

decompose or unblend image into layers to make them easier to edit . it would be useful to fix small details . 7
here is a project that allows decompose images into layers but its a little bit broken but the cli option can be easily fixed . https://github.com/yuki-koyama/unblending gui

Proposed workflow

it would like when a generated image is send to img2img option .

Additional information

No response

78Alpha commented 1 year ago

Seems to be a PoC? Has no mention of memory usage but the web demo forces all images into a 256x256 square, so a lot of information would be lost, as well as non-square images being warped.

Of course, it also lacks testing on windows native. Judging by the complexity, it will likely be told to be made an extension.

The web demo does have some good results, it manages to break it up into expected layers and make the final image again.

asecino1999 commented 1 year ago

it doesn't use more than 1.2 gb of ram ( for a image of 2k x 3k pixels ) but it doesn't use the gpu also the time grows up very fast , it rook about 2 hours to process an image of 2k x 1k pixels . i could run it in a colab too https://colab.research.google.com/drive/1Znida83nM2946TjM0ZmCN2waLjeFNeCg?usp=sharing