MewZoom 4X (super-resolution)
andrewdalpino/MewZoom-4X is a super-resolution model that targets a common “real-world” upscaling need: not just increasing resolution, but also removing blur, noise, and compression artifacts along the way. The project is packaged like a small library (the ultrazoom Python package), and the model card is unusually concrete about how to load weights, run inference, and reproduce its examples.
What makes it worth a look is the combination of deployment ergonomics and controllability. In addition to standard 2×/3×/4×/8× checkpoints, the project includes “control” variants where you pass a small vector describing the expected degradation (blur/noise/JPEG). That gives you a useful knob for dialing down over-sharpening on clean inputs, or pushing harder on low-quality images. A good first test is to run the non-control and control versions on a handful of your own images (especially ones with visible compression artifacts) and see whether the controlled enhancements improve perceived detail without inventing textures.
Quick stats from the listing feed: pipeline: image-to-image · 9 likes · 17 downloads.
Source listing: https://huggingface.co/models?sort=modified