Yellowed, scratched, or faded family photos always seemed doomed to be forgotten. But with the advance of convolutional neural networks and diffusion models, restoring these images is now within anyone's reach — no need to pay a graphic designer or learn Photoshop.

In this guide, you'll learn about the best AI tools available in 2026 for restoring old photos, with an honest comparison of features, limitations, and who each one is best suited for.

What can AI do to old photos?

Modern photo restoration models can perform several tasks automatically or semi-automatically:

💡 Tip: the better the original scan, the better the results will be. If possible, scan the photo at 600 DPI before processing it with AI.

Comparison of the main AI restoration tools

ToolStrengthsFree?Ideal for
ReminiFacial restoration, mobile appPartialSelfies and portraits
MyHeritage Photo EnhancerColorization + animationPartialFamily photos
Adobe FireflyAdvanced inpainting, scratchesNoProfessionals
Stable Diffusion (img2img)Full control, freeYesAdvanced users
Palette.fmArtistic colorizationYesColorizing B&W

Remini: the most popular for faces

Remini is probably the best-known tool for restoring photos of people. Its model was trained specifically on faces, which results in impressive improvements to old portraits — recovering eyes, hair, and expressions that seemed lost forever.

The free plan allows a few restorations per day. For heavy use, the paid plan offers unlimited processing. The app is available for Android and iOS, making the process quite convenient.

MyHeritage: the choice for genealogy

MyHeritage Photo Enhancer goes beyond simple restoration: it offers automatic colorization and even an "animation" feature that makes the faces in the photo move. For anyone wanting to preserve family memories, it's one of the best feature combinations available. The colorization result is surprisingly natural, especially for outdoor scenes with natural light.

Stable Diffusion: full power at no cost

For users who want maximum control and don't mind a steeper learning curve, Stable Diffusion with models like RealESRGAN and GFPGAN offers professional-quality results at no cost at all. You can combine upscaling, facial restoration, and colorization in a single workflow.

💡 Tip: after restoring the photo with AI, use ImageTools' Image Compressor to reduce the file size without losing the quality you gained.

How to get even better results

Regardless of the tool you choose, a few practices improve the quality of the final restoration:

  1. Physically clean the photo before scanning it (use a dry microfiber cloth)
  2. Scan at high resolution (minimum 600 DPI for 4×6 in photos)
  3. Save as TIFF or PNG so you don't add JPG compression before processing
  4. Try more than one tool — each model has different strengths
  5. Finish by manually adjusting brightness and contrast, if needed

Improve the quality of your restored photo

After restoring, use ImageTools to convert, compress, or resize the image without losing the detail the AI recovered.

Compress image now

Frequently Asked Questions

Can AI restore severely damaged photos?
It depends on how bad the damage is. Surface scratches, fading, and grain are easily fixed. Large tears or completely missing areas are harder — tools with inpainting (like Firefly or Stable Diffusion) get better results in those extreme cases.
Is automatic colorization faithful to the original colors?
Not necessarily. AI infers colors based on learned patterns — it doesn't "know" the actual color of the clothing or setting. For historically precise photos, consult period references. For personal, sentimental use, the results tend to be very convincing.
What's the best format to save the restored photo in?
Use PNG to preserve maximum quality with no compression. If you need a smaller file to send or share, convert to JPG at 90% quality or higher — ImageTools has a free converter for that.
Do I need a powerful computer to use these tools?
For online tools like Remini and MyHeritage, no — everything runs in the cloud. For local Stable Diffusion, a dedicated GPU (like an NVIDIA card with 6GB+ VRAM) is recommended for speed, but there are versions that run for free via Google Colab.