amazing modification to k_euler resynthesizing original image in img2img!

[1blackbar]

Can You actually add this to the webuit ? he made a code already and made it public
https://www.reddit.com/r/StableDiffusion/comments/xboy90/a_better_way_of_doing_img2img_by_finding_the/
Hes resynthesizing original image perfectly for editing !
Heres the code but more about it is on reddit post
https://gist.github.com/trygvebw/c71334dd127d537a15e9d59790f7f5e1

[oobabooga]

Amazing!!!

[AUTOMATIC1111]

see the scripts for initial version of this

[Guguskis]

Could someone add instructions how to properly use this img2imgalt script? I mean like what options to select in image2image section and what configuration to tweak

[AUTOMATIC1111]

It's in the readme.

[lcubeddu]

The generated image gets weirdly noisy/high contrast in the last few iterations (no matter the iteration count)
Here is an example.

Original Prompt : White duck, dark background, flat design
Img2Img Prompt : white duck, blue background, flat design
Decode CFG Scale : 0.7
Decode steps : 50

Iterations : 50
CFG Scale : 2

Original duck:

During the generation:

Finished image:

Using 25 iterations yields the same results: last few iterations up the contrast.
Doesn't happen with the image of the original post, i guess it's just a downside of the technique

[DavidePaglieri]

Hi, I get the following error when trying out the img2img alternative:

TypeError: expected Tensor as element 0 in argument 0, but got ScheduledPromptBatch

[Blucknote]

Hi, I get the following error when trying out the img2img alternative:

TypeError: expected Tensor as element 0 in argument 0, but got ScheduledPromptBatch

fixed in 7fe00d0

[C43H66N12O12S2]

duplicate of #288