SUPIR
SUPIR Intelligent Image Restoration, created by XPixel Group, is a large-scale AI tool designed to reimagine and restore images by blending...
Last verified:
What is SUPIR?
SUPIR (Scaling-UP Image Restoration) is a high-fidelity general image restoration model based on large-scale diffusion generative prior technology. It provides text-driven, intelligent restoration that blends AI technology with creativity to give every image a brand new life. The tool excels at enhancing and restoring low-quality images to stunning, photo-realistic quality.
Key features include landscape image restoration that highlights natural beauty, ultra-resolution for faces revealing true expressions with lifelike clarity, animal image restoration bringing fur and charm to life, gaming image enhancement for remastered worlds in stunning detail, cinematic rebirth restoring classic films in pristine detail, and vintage photo revival bringing history to life with vivid clarity. The tool supports various restoration tasks across different image genres.
SUPIR is designed for photographers looking to enhance their work, designers needing high-quality image restoration, gamers wanting to remaster gaming visuals, historians and archivists restoring vintage photographs, film enthusiasts rebirthing classic cinematic works, and anyone looking to enhance or restore degraded or low-quality images with high fidelity and clarity.
The model offers two versions: SUPIR-v0Q for balanced quality versus speed, and SUPIR-v0F for highest fidelity at the cost of extra compute. It includes parameters for upscale multiplication, diffusion steps, text prompt guidance strength, and control scale to retain input image consistency.
SUPIR pricing
Pricing model: Free
SUPIR uses a Freemium pricing model, meaning there is a free tier along with other paid options. Specific pricing details for paid plans are not publicly disclosed on the website. The tool is free to use through research implementations and repositories. The official commercial product based on SUPIR technology is SupPixel AI at suppixel.ai which offers the next level of image processing and upscaling.
SUPIR pros
- Text-driven intelligent restoration process
- High-fidelity general image restoration model
- Large-scale diffusion generative prior technology
- Restores landscape photos with natural beauty enhancement
- Ultra-resolution capability for face detail enhancement
- Reveals true expressions in portraits with lifelike clarity
- Animal image restoration bringing fur and charm to life
- Gaming image enhancement for remastered stunning clarity
- Cinematic rebirth restoring classic films in pristine detail
- Vintage photo revival bringing history to vivid clarity
- Two model versions: Q for speed, F for highest fidelity
- Supports upscale multiplier from 1x to 4x resolution
- Text prompt guidance with positive and negative hints
- Control scale parameter to retain input image consistency
- Available through GitHub repository with documentation
- Works with batch jobs and single-shot runs
- Suitable for historical B&W enhancement and colorization
- Noise reduction for low-light and noisy shots
- Super-resolution upscaling capabilities
- Photo-realistic restoration quality output
SUPIR cons
- No live online demo currently available
- Documentation only available on GitHub
- Requires high resource availability for processing
- No official API availability mentioned
- No batch processing interface for beginners
- No offline desktop version available
- Limited by input image quality
- Research paper details not fully public
- No customer support mentioned
- Complex parameter configuration for beginners
- Free tier limitations not clearly disclosed
- Paid plan pricing not publicly disclosed
- Requires conda environment setup for local use
- Checkpoints must be downloaded from HuggingFace
- Slower processing with SUPIR-v0F high fidelity mode
Frequently asked questions about SUPIR
What is SUPIR and what does it do?
SUPIR (Scaling-UP Image Restoration) is a groundbreaking image restoration method that harnesses generative prior and the power of model scaling up. It is a text-driven, intelligent restoration tool that blends AI technology with creativity to give every image a brand new life. The tool excels in enhancing and restoring low-quality images to high-fidelity, photo-realistic quality across various genres including landscapes, portraits, animal photos, gaming visuals, cinematic frames, and vintage photographs.
What types of images can SUPIR restore?
SUPIR can restore various types of images including landscape photos (highlighting natural beauty), portraits and faces (revealing true expressions with ultra-resolution), animal pictures (bringing fur and charm to life), gaming images (creating stunning clarity and detail), cinematic frames and classic films (restoring in pristine detail), and vintage photographs (bringing history to life with vivid clarity). It works on any degraded or low-quality image.
What are the two SUPIR model versions and which should I choose?
SUPIR offers two model versions: SUPIR-v0Q (default) which provides balanced quality versus speed, and SUPIR-v0F which offers the highest fidelity at the cost of extra compute time. Choose SUPIR-v0Q for faster results with good quality, or SUPIR-v0F when you need the absolute best quality and can wait longer for processing.
How does the text-driven restoration work in SUPIR?
SUPIR employs a text-driven restoration process where you provide positive prompts (describing what you want the image to look like, such as 'photo realistic, vibrant color, super detailed') and negative prompts (what to avoid, such as 'noise, blur, artifacts'). The s_cfg parameter controls guidance strength for text prompts, with higher values enforcing prompts more strictly. Better prompt descriptions produce better restoration results.
What upscaling capabilities does SUPIR offer?
SUPIR supports upscale multiplication where 1 = no change, 2 = 2x resolution, and up to 4x resolution. The upscale parameter is an integer multiplier for output resolution. For super-resolution tasks, users can set upscale to 2 or 4 based on their target resolution, and increase edm_steps to 100 for extra detail.
How do I access and use SUPIR?
SUPIR can be accessed via its official website at supir.xpixel.group or through the GitHub repository. For local use, clone the repository from GitHub, navigate to the SUPIR directory, create a conda environment, install requirements from requirements.txt, download necessary checkpoints from HuggingFace, and run inference using gradio_demo.py for interactive options or on validation datasets. An online demo platform is coming soon.
Can SUPIR restore historical black and white photos?
Yes, SUPIR is specifically designed for historical B&W enhancement. For this task, use use_llava=true to auto-generate context captions, set color_fix_type to Wavelet, use positive prompts like 'vivid colorization, realistic skin tones, natural lighting', negative prompts like 'over saturation, unnatural hues', and set edm_steps to 60 for optimal historical photo restoration with natural colorization.
What parameters should I adjust for low-light or noisy images?
For low-light and noisy shots, increase edm_steps to 75 for more diffusion steps, set s_noise to 1.005 (slightly higher for robust denoising), set s_churn to 4 to stabilize heavy noise removal, use positive prompts like 'clean, sharp, high fidelity', and negative prompts like 'grain, excessive noise, blur'. These settings optimize SUPIR for denoising challenging low-light photographs.
What are the system requirements for running SUPIR locally?
SUPIR requires high resource availability, especially when using higher resolutions or complex restoration tasks. You need a powerful machine with sufficient GPU memory for running the large-scale diffusion model. For systems lacking necessary resources, consider using a cloud solution like Think Diffusion. Local setup requires conda for environment management and access to HuggingFace for checkpoint downloads.
Is there an API available for SUPIR and can I use it programmatically?
SUPIR is available through the Segmind API which provides a Photo-Realistic Image Restoration API. You can launch batch jobs or single-shot runs via the Segmind API, HuggingFace, or GitHub integration programmatically. The API supports CLI and Python SDK for batch scripting to iterate parameters programmatically and identify optimal configurations for your restoration needs.