zsyOAOA/ResShift

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

1427 stars · 95 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

ResShift is an efficient diffusion model for image super-resolution that transfers between low- and high-resolution images by shifting residuals, achieving strong results with as few as 15 sampling steps. It includes training and inference code, plus Gradio, Colab, and Replicate demos, and also supports deblurring.

Use cases

• upscale low-resolution images with a diffusion model
• fast image super-resolution with few sampling steps
• real-world image super-resolution
• deblur blurry photos with a diffusion model
• train a custom super-resolution diffusion model
• compare super-resolution models on synthetic and real datasets

When to choose

• you need high-quality super-resolution with fast inference
• you want a research-grade diffusion SR model with training code
• you need both super-resolution and deblurring in one framework

When to avoid

• you need a lightweight non-deep-learning upscaler for production apps
• you lack a GPU and cannot afford diffusion inference costs
• you need a simple GUI-only tool without coding

Facets

library · maturity active

image-processing machine-learning deep-learning image-processing computer-vision machine-learning artificial-intelligence python cross-platform diffusion-model super-resolution image-restoration deblurring neurips research-code gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="zsyOAOA/ResShift")