Image Upscaling GUI based on ESRGAN
Cupscale hasn't been updated for ~1 year at the time of forking, but still has a lot of unexpected behavior/interface bugs that that don't really make it unusable but do make it a bit annoying to deal with for daily use. I probably won't fix many deeper implementation issues outside of the RTX 3060 CUDA detection bug, but things like Cupscale throwing when a preview image isn't available even when switching to a different tab, would be worked on.
This fork will be dormant while I study the repo to ensure I understand what's going on before I change things around and break stuff. It's also more of a personal use fork so I won't make any guarantees on the extent to which I fix non-usability-breaking bugs.
Based around xinntao's ESRGAN implemented via Joey's Fork.
AMD/Intel GPU compatibility is possible thanks to BlueAmulet's esrgan-ncnn-vulkan based on nihui's realsr-ncnn-vulkan running on Tencent's ncnn framework, as well as xinntao's Real-ESRGAN.
The application is more or less portable. It's a single executable that you can run anywhere.
Temporary files are stored in the installation directory by default, which is why you shouldn't install the application in protected locations like Program Files.
- Nvidia CUDA (Recommended)
- Vulkan (Works on any modern GPU, but is slower and takes a long time start up)
- CPU (Works without GPU, but is very slow)
- CUDA, Vulkan/NCNN or CPU supported, with included model converter for NCNN
- Model Interpolation
- Model Chaining (Run images through multiple models at once)
- Batch Upscaling (Load a directory or multiple single images)
- Automatic Image tiling/merging to avoid running out of VRAM
- Pre-Processing: Optionally downscale images before upscaling
- Post-Processing: Automatically resize after upscaling
- Compatible with PNG, JPEG, BMP, WEBP, TGA, DDS images
- Loading an image from the clipboard (no need to download images from web)
- Various types of comparisons (Side-By-Side, 50/50, and before/after animations as GIF or MP4)