Description
This is my latest anime model. It's a continuation of what I tried to do with AnimeSharpV3: faithful upscaling with as few artifacts as possible.
I have included two versions. The base RCAN model, then the Fast model. The base model will work great on just about anything, and the fast model is probably ~95% of the quality but much faster to run 😊
Here are some comparisons:
Note
If you intend to use this as a normal upscaling model, you'll want the latest version of chaiNNer-nightly: https://github.com/chaiNNer-org/chaiNNer-nightly
If you want to use this model with diffusion, you'll need to use the latest ComfyUI, and upgrade the spandrel python library to 0.4.1. This will add support for the architecture (RCAN)
Description
V4 release
FAQ
Comments (5)
Hello, any planning to make it with ESRGAN .pth version? look like RCAN doest support on A1111, yes there already folder named RCAN but im not sure its works because i still cant use it on A1111.
I'll look into it, but really I'd rather not. V3 was ESRGAN, and it was pretty limited in what I could do at reasonable training settings. RCAN allowed me to get higher quality results while having faster output (for the Fast model at least).
I'll reply here if I end up releasing one
a1111 comes only with spandrel 0.3.4 which too old. As mentioned in the description at least 0.4.1 is needed.
You could try to upgrade the installed version of it in your python environment.
pip install spandrel -U
pip install spandrel-extra-arches -U
 Ah, I wasn't aware that A111 uses spandrel now. That should work :)
For some reason the site doesn't let us put this in the Resources section of our posts, but I just wanted to say thank you and that I use this on everything I generate.