Try Seedance 2.0 from ByteDance
Collections

Restore images

These models restore and improve images by fixing defects like blur, noise, scratches, and damage.

Models we recommend

Best all-around: Restore Image (FLUX Kontext)

Restore Image handles the most common restoration tasks in one model — fix scratches and damage, remove artifacts, and colorize black-and-white photos. Describe what you want fixed in plain English and it handles the rest. The easiest starting point for most restoration work.

Best face restoration: GFPGAN

GFPGAN is still the go-to for restoring faces. It runs in a fraction of a second and produces natural-looking results on old photos and AI-generated faces. For even more realistic faces with better identity preservation, try CodeFormer.

Best colorization: DDColor

DDColor adds vibrant, realistic color to black-and-white photos. It handles portraits, landscapes, and even anime-style images well. For multiple color variations from one input, try BigColor.

For general deblurring and denoising: NAFNet

NAFNet cleans up noise and blur without over-smoothing. A good complement to face-specific models when you need to improve the entire image.

For old damaged photos: Bringing Old Photos Back to Life

Microsoft's Bringing Old Photos Back to Life is specifically designed for heavily damaged photos with scratches, tears, and fading. It repairs physical damage while preserving the original content.

Tips

  • For best results on old photos, restore faces first with GFPGAN or CodeFormer, then upscale with a model from our upscaling collection →
  • Modern image editing models like Nano Banana and GPT Image 1.5 can also restore photos conversationally — describe what's wrong and they'll fix it
  • Colorization works best when the base image is clean — restore first, then colorize

Recommended Models

Frequently asked questions

Which model should I start with?

flux-kontext-apps/restore-image handles the most common restoration tasks — scratches, damage, artifacts, and colorization — in one model. Describe what's wrong in plain English and it fixes it. For face-specific restoration, use tencentarc/gfpgan.

Which models are the fastest?

tencentarc/gfpgan runs in a fraction of a second for face restoration. piddnad/ddcolor is fast for colorization. flux-kontext-apps/restore-image takes a few seconds but handles multiple tasks at once.

What works best for restoring old photos with faces?

Start with tencentarc/gfpgan for face restoration — it reconstructs facial features realistically even from blurry or degraded originals. For better identity preservation, try sczhou/codeformer. After restoring faces, upscale with a model from the upscaling collection.

What's best for colorizing black-and-white photos?

piddnad/ddcolor produces vibrant, realistic colorization across portraits, landscapes, and even anime. For multiple color variations from one input, try cjwbw/bigcolor. You can also use flux-kontext-apps/restore-image with a prompt like "colorize this black and white photo."

Can modern image editing models also restore photos?

Yes — conversational image editing models like google/nano-banana and openai/gpt-image-1.5 can restore photos when you describe the damage. They're good for complex or unusual damage patterns that specialized restoration models might not handle well.

What's the best workflow for heavily damaged photos?

  1. Use microsoft/bringing-old-photos-back-to-life to repair scratches, tears, and physical damage.
  2. Restore faces with tencentarc/gfpgan or sczhou/codeformer.
  3. Colorize with piddnad/ddcolor if the photo is black and white.
  4. Upscale with topazlabs/image-upscale or recraft-ai/recraft-crisp-upscale.

Can I use these models commercially?

Most models support commercial use, but check the license on each model page. Some academic models may have research-only restrictions.