> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zevor.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Restore an old photo

> Bring back an old, damaged or black-and-white photo with AI: repair scratches, colorize and upscale in a couple of steps. Recipe with exact credits.

Restoring an old photo chains two tools: **Restore** (repairs and
optionally colorizes) and **Upscale image** (bumps resolution to a
print-ready size). It takes 2-3 minutes.

## Outcome

* 1 restored photo (repaired + colorized if it was B\&W).
* Optional: same photo upscaled 2x-4x for printing or archival.

## Time and credits

* **Total time**: \~2-5 minutes.
* **Credits**: 4-8 cr.

A \$4.99 top-up pack (30 cr) restores \~5-7 photos.

## Steps

<Steps>
  <Step title="Upload the original">
    At [zevor.ai/restore-old-photo](https://zevor.ai/restore-old-photo),
    drop the file. JPG/PNG, up to 5 MB. If it's larger, open it in
    Preview or Photos and export at medium quality before uploading.
  </Step>

  <Step title="Restore">
    Generate. The AI repairs creases, scratches, halos and blur, and
    colorizes if the photo was black and white.

    Cost: 4 cr.
  </Step>

  <Step title="Upscale (optional)">
    If you want to print or archive in high resolution, run it through
    [Upscale image](https://zevor.ai/ai-image-upscaler) at 2x.

    Cost: 4 cr.
  </Step>

  <Step title="Download">
    **Download** button under the result.
  </Step>
</Steps>

## Best practices

* **Scan before photographing**. A photo of a photo loses detail. If
  you have a scanner, scan at 600 dpi minimum and upload the JPG.
* **Crop before uploading** if there's a frame, mat or album around
  the photo: the AI can confuse those edges with part of the image.
* **One photo at a time**. If you have a sheet with several, crop
  them into separate files.

## Common mistakes

* **Uploading a faceless photo expecting facial reconstruction**: the
  model restores what it sees. If the face is fully destroyed, it
  fills with its best estimate, not with the actual person.
* **Drastic, unrealistic changes**: if the photo is badly damaged,
  the result can vary across attempts. Generate 2-3 times and keep
  the best.

<Note>
  For old low-resolution videos, see the
  [Upscale video](/en/video-tools/upscale-video) recipe.
</Note>