stable-diffusion-webui
stable-diffusion-webui copied to clipboard
multiple trivial changes for "extras" models
This is a small PR is to add a few requests regarding the "extra" models available. Most of the changes are relatively trivial and don't impact the application logic. These are the changes:
- added support for realESRGAN architecture with additional up_conv block ("8x") (https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/3536).
- automatically move BSRGAN models from the previous BSRGAN directory to the ESRGAN directory, to avoid BSRGAN model from "disappearing" when updating the code. (https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/3551, https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/3596).
- added basic Nearest Neighbor upscaling option when working with pixel-art and avoid blurry scaling (Discord request).
- change maximum upscaling size to 8, to accomodate for larger upscale factor models.
- set the default upscaling value to 4, the "native" upscaling factor of most models.
- allow use of 1x scale models like ESRGAN, ScuNET, and others that are mainly used for denoising and deblurring without changing images scales. (https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/3526).