Update on the banding guide

[Defaltsiuncula]

Hi 👋 I worked on small changes for the banding guide 4 days ago when I was working on the ER guide, but the PR was postponed. The reason for that is that I checked ESGRAN. I thought about adding that to the guide, but only to cover the banding part, of course. And that would be the second part of the guide, after Lightroom. After checking issues on GitHub, I saw a post (moyevka) where you want a guide to be around those models.

Is it fine for you guys to cover the ESGRAN there? I think it would be wise to connect ESGRAN models to the banding guide. Since it's already mentioned, it would just explain further how to use it, etc., and it would be around debanding models, not dithering, anti-aliasing, etc. since of course, the guide is about how to solve color banding quickly. I usually do stuff before asking and just deliver, but I thought for this one it's better to consult. It is completely understandable if you would want ESGRAN out of the banding guide so it has its own complete guide, but let me know what you think

[Defaltsiuncula]

I wanted to close this issue, but I reopened it to give some feedback on the ESGRAN. IDK if it's the type of image or what, but I don't dig the results entirely from this. I tried all deband x1 models. Only one of the models did okay all around namely 1x_N64clean.pth, but you can still see lines (path) where the banding was actually in the sky. x1_ITF_SkinDiffDDS_v1.pth did okay on some parts of the sky, but on some parts, you could still spot some banding.

It's easy to spot, though adding a bit of noise/grain would solve that. With other models, some parts of the image are 'destroyed'. It also takes time for the AI to do this (changing the tile size affects that). Well, at least when the image is 6k res, so I can't characterize that as an issue.

For more tests, images are compressed further via FastStone with a 50% reduced size. I guess everything is correctly set.

Here is ChaiNNerSetup, x1_ITF_SkinDiffDDS_v1 result, x1_Bandage-Smooth-[64] result, x1_Debandurh-FS-Ultra-Life, x1_N64clean, and vanilla image. It would be best to download these images and put them into the folder to compare.