An old family photo can be faded, scratched, creased, or blurry. AI image tools can reduce these problems and bring back contrast and detail. They can also invent detail that was never in the photo. This guide explains how to prepare an old photo, what to ask for, and how to check the result so you keep what is real.
What AI restoration can and cannot do
| Usually works well | Needs care |
|---|---|
| Removing small scratches, dust, and stains | Large missing areas, where the tool must invent content |
| Restoring faded contrast | Faces that are very small or very blurry |
| Reducing noise and mild blur | Adding color to black and white photos, which is a guess |
| Straightening and cleaning edges | Text, badges, and signs, which may be rewritten incorrectly |
The result is an interpretation of the photo, not a perfect recovery. Keep the original file.
Step 1: Make a good scan or photo
- Use a flatbed scanner at 600 dpi or higher if you have one. Otherwise photograph the print in daylight, straight on, without glare.
- Clean the glass and the print surface gently, without liquids.
- Crop to the photo, with no table or fingers in the frame.
- Save a copy as a lossless or high-quality file.
Step 2: Be careful with private photos
Old family photos often show people who cannot give consent now. Before you upload a photo to an online AI tool, read how the tool stores and uses images. Share only photos you are comfortable sending to that service.
Step 3: Ask for repair, not reinvention
The more you ask the tool to change, the more it may invent. Start with a conservative prompt.
Restore this old photograph. Remove scratches, dust, and stains, and repair fading. Keep the people’s faces, expressions, clothing, and the composition exactly the same. Do not add or remove people or objects. Keep the original black and white tones.
Step 4: Decide about color
Adding color to a black and white photo is a creative choice, not a restoration. The tool guesses clothing, hair, and skin colors. If you add color, treat it as an interpretation, and keep the black and white version too.
Add natural, muted colors to this black and white photograph. Keep faces, expressions, and the composition exactly the same. Use realistic skin tones and subdued clothing colors.
Work in two passes
Restoration and enhancement are two different jobs, and doing them in one prompt makes it hard to see what the tool changed. Work in two passes instead:
- Repair pass: remove scratches, dust, stains, and fading. Do not ask for color, sharpening, or a new look.
- Enhance pass: only if you want it, add gentle sharpening or colorization to the repaired image.
Save the output of each pass as its own file. If the second pass goes wrong, you can return to the clean repaired version.
Prepare different types of damage
| Damage | What to do first |
|---|---|
| Fold or crease line | Scan flat, with no shadow along the crease, and say “remove the fold line” |
| Water stains | Ask to “remove stains and keep the original tones” |
| Fading | Ask to “restore contrast and tonal range” |
| Torn corners | Crop to the intact area first, rather than letting the tool invent the missing piece |
| Small face, low detail | Use a larger scan, and accept that detail may be a guess |
Document what you changed
If you plan to share the restored image with family or in an archive, add a short note about the work. A simple line works: “Scanned from the original print in 2026 and restored with AI tools. Colors added by AI.” This helps relatives later, and it keeps a clear difference between the record and the interpretation.
Know when to stop
Each additional pass moves the image further from the original. If the photo already shows faces and the main details clearly, a light cleanup is often better than a full rebuild. Stop when the people look like themselves and the photo still feels like it came from its time.
Step 5: Check the result carefully
- Compare faces with the original. Look at the eyes, nose, and mouth of each person.
- Check hands, hair, and jewelry for invented detail.
- Look for text on buildings or clothing that changed.
- Check the background for new objects.
- Ask a family member who knew the person whether the face looks right.
Common problems and fixes
| Problem | Fix |
|---|---|
| A face looks younger or different | Ask for “keep the face exactly as in the original”, and use a larger scan |
| The photo looks too smooth | Add “keep natural film grain and skin texture” |
| Colors look artificial | Ask for “muted, natural colors”, or keep it black and white |
| Added or removed objects | State “do not add or remove anything” and run again |
Frequently asked questions
Should I replace the original with the restored image?
No. Keep both. The restored file is an interpretation, and the original is the record.
Can AI restore a photo that is badly torn or missing a face?
It can fill the gap, but it will invent the missing part. Treat that area as a guess and be open about it if you share the image.
What resolution should I scan an old photo at?
600 dpi is a common choice for prints, and higher is useful for very small photos. Larger files give the tool more detail to work with, but check the upload limit of the tool.
Is it right to colorize a photo of a relative?
It is a personal and creative choice. Remember the colors are guesses. If you know the real colors, such as from a family story, describe them in the prompt.
For prompt structure and wording, read how to write a good AI image prompt. To keep faces unchanged, see how to keep your face the same in AI photo edits. For tone and contrast choices, read the black and white portrait guide.
