image/video super-resolution and restoration
Real-ESRGAN is a PyTorch-based practical image/video super-resolution and restoration tool using ESRGAN-type GAN models, with pre-trained models for real-world and anime content.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
Transformer-based restoration/SR alternative; choose when you want high PSNR on known degradations without GAN artifacts.
SwinIR: Image Restoration Using Swin Transformer (official repository)
Diffusion-based photorealistic restoration; choose for top perceptual quality when you have large GPU resources.
SUPIR aims at developing Practical Algorithms for Photo-Realistic Image Restoration In the Wild. Our new online demo is also released at suppixel.ai.
Recent diffusion-style real-world super-resolution method; choose for state-of-the-art quality at higher compute cost.
(CVPR 2025) Adversarial Diffusion Compression for Real-World Image Super-Resolution [PyTorch]
Universal image restoration with language-controllable degradation; choose when you need one model for many tasks.
[ICLR 2024] Controlling Vision-Language Models for Universal Image Restoration. 5th place in the NTIRE 2024 Restore Any Image Model in the Wild Challenge.
High-capacity diffusion-based real-world restoration model; choose for strong perceptual restoration with heavy compute.
[NeurIPS 2024] DreamClear: High-Capacity Real-World Image Restoration with Privacy-Safe Dataset Curation
SDE-based general image restoration approach; choose when a diffusion-like formulation fits your research or use case.
Image Restoration with Mean-Reverting Stochastic Differential Equations, ICML 2023. Winning solution of the NTIRE 2023 Image Shadow Removal Challenge.
Transformer super-resolution with hybrid attention; choose for high-PSNR restoration on standard degradations.
CVPR2023 - Activating More Pixels in Image Super-Resolution Transformer TPAMI - HAT: Hybrid Attention Transformer for Image Restoration
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
Dedicated blind face-restoration model; choose when face identity matters more than general scenes.
[NeurIPS 2022] Towards Robust Blind Face Restoration with Codebook Lookup Transformer
Specialized real-world face restoration; choose for old photos where faces are the main subject.
GFPGAN aims at developing Practical Algorithms for Real-world Face Restoration.
Lightweight efficient SR model; choose for edge/low-compute deployment, not for blind real-world degradation.
[ICCV 2023] Spatially-Adaptive Feature Modulation for Efficient Image Super-Resolution; runner-up method for the model complexity track in NTIRE2023 Efficient SR challenge
Dedicated blind JPEG artifact removal; choose when compression artifacts are the primary degradation.
Official Code for ICCV 2021 paper "Towards Flexible Blind JPEG Artifacts Removal (FBCNN)"
Older multi-task restoration CNN; choose if you prefer a non-GAN baseline for denoising/SR/enhancement.
[ECCV 2020] Learning Enriched Features for Real Image Restoration and Enhancement. SOTA results for image denoising, super-resolution, and image enhancement.
Anime-specific real-world super-resolution; choose for anime images/videos where specialized training pays off.
APISR: Anime Production Inspired Real-World Anime Super-Resolution (CVPR 2024)
Transformer SR specialized for compressed inputs; choose when JPEG-like compression is the main issue.
[ECCV] Swin2SR: SwinV2 Transformer for Compressed Image Super-Resolution and Restoration. Advances in Image Manipulation (AIM) workshop ECCV 2022. Try it out! over 3.3M runs https://replicate.com/mv-lab/swin2sr
Predecessor blind SR GAN; choose for a simpler, older baseline instead of the more refined Real-ESRGAN.
Designing a Practical Degradation Model for Deep Blind Image Super-Resolution (ICCV, 2021) (PyTorch) - We released the training code!
Classic anime-style CNN upscaler; choose for a lightweight, legacy alternative specifically for anime art.
Image Super-Resolution for Anime-Style Art
Does the same job, but ships as an app. Useful to a person, not swappable into a codebase.
End-user GUI/executable for AI image upscaling; choose when you want a turnkey app rather than a Python library.
AI超分辨率工具
Real-time shader-based anime upscaler for video players; choose for live playback rather than offline restoration.
A High-Quality Real Time Upscaler for Anime Video
End-user desktop app for AI image/video upscaling; choose for a friendly GUI rather than a programmatic API.
QualityScaler - image/video AI upscaler app
These projects were analysed and named Real-ESRGAN among their alternatives. The relationship is not symmetric — how Real-ESRGAN rates them is a separate judgement, made when Real-ESRGAN is analysed in its own right.
calls Real-ESRGAN “choose when restoring whole images/video rather than only faces; no face-specific identity prior”
GFPGAN aims at developing Practical Algorithms for Real-world Face Restoration.
calls Real-ESRGAN “Choose for practical general-purpose image/video restoration with a focused, widely used implementation.”
OpenMMLab Multimodal Advanced, Generative, and Intelligent Creation Toolbox. Unlock the magic 🪄: Generative-AI (AIGC), easy-to-use APIs, awsome model zoo, diffusion models, for text-to-image generation, image/video restoration/enhancement, etc.