SUPIR
Scaling-up image restoration: photo-realistic upscaling trained on 20 million high-resolution images.
About
SUPIR (Scaling-UP Image Restoration) is a diffusion-based upscaler built on Stable Diffusion XL, trained on 20 million high-resolution images with descriptive text annotations. Unlike classic super-resolution it restores images guided by detailed positive and negative text prompts, which lets you steer the regeneration — from fidelity-focused cleanup to creative enhancement. The official repo ships inference code and pruned checkpoints, and there is a well-maintained ComfyUI integration.
It demands real hardware (24 GB+ VRAM for comfortable use) and the license is research-only: non-commercial use, with a contact for commercial inquiries. With ~5,700 GitHub stars it is the go-to diffusion upscaler when a regressive model like Real-ESRGAN under-fills detail, but its generative nature means it invents texture — great for photos, wrong for logos and text.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Provider-reported (paper, arXiv 2401.13627, CVPR 2024): evaluated on real-world photo restoration benchmarks with strong fidelity-quality trade-offs per authors
- No independent third-party benchmark replications published
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (TIER verified, metrics-v1). Evidence: CVPR 2024 paper (arXiv 2401.13627); ~5,700 GitHub stars; widely adopted in ComfyUI/Forge upscaling pipelines. Strengths: largest-scale training of any open restorer, prompt-steerable, heavy real-world adoption. Docked for: (1) non-commercial research-only license — commercially unusable without permission; (2) eval claims author-reported. License: custom non-commercial research license (no SPDX identifier).
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 5.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · CVPR 2024 paper (arXiv 2401.13627); ~5,700 GitHub stars (verified 2026-10-16)
- · Widely adopted in ComfyUI/Forge upscaling pipelines
- · Non-commercial license - commercially unusable without author permission