Very experimental - still training. (made a v2 from scratch with different settings and higher dimensions but its identical to this one plus I've discovered many limitations I was not aware of so I wont share it as its not worth it and I'm ending this chapter as a failed experiment)

Triggerword: uncensor

Trained on just about any type of image: sketch, monochrome, comic, anime, hentai and realistic images

Trained to handle: white/black bar censors, mosaic censoring and gaussian blur censoring - on penis, pussy, spread pussy, anus and nipples

Limitations:

• slight color mismatch

• cant handle very strong mosaics

• does not handle bar censoring well due to color mismatching in colored inputs and for some reason it also freaks out to do the same in monochrome inputs

• despite being trained on a very diverse dataset - it seems mostly only useful for light mosaic decensoring in colored inputs

Requires the provided custom workflow and the following custom nodes:

https://github.com/kijai/ComfyUI-KJNodes (image resizing node)

https://github.com/Comfy-Org/Nvidia_RTX_Nodes_ComfyUI (for fast, lightweight and good upscaling - replace it with something else if you want)

https://github.com/GreenLandisaLie/ComfyUI-Anima-LLLite-Inpaint (a fork from the original 'ComfyUI-Anima-LLLite' node by kohya. this fork allows you to specify the strength of the white part of the mask)