Quick answer
Descratch & Repair removes scratches, dust, creases, and tears from scanned photographs using the Evoke AI mobile app for Android and iOS. It works by detecting damage and filling it with surrounding texture. Small and medium damage disappears convincingly; where a tear has destroyed a face or a large area, the model reconstructs something plausible rather than what was actually lost.
What Damage Can Actually Be Repaired?
Photo repair works by inpainting: the model identifies pixels that look like damage rather than image content, then fills them using the texture around them. How well that works depends almost entirely on how much undamaged material surrounds the damage.
A hairline scratch across a wall has plenty of context — the fill is invisible. A crease running through someone's eye has almost none, so the model has to invent an eye. It will produce something anatomically sensible, but it is a reconstruction.
- Repairs cleanly: dust specks, fine scratches, small spots, mildew flecks, scanner lines.
- Repairs well: creases across backgrounds, clothing, and hair; torn corners; edge damage.
- Reconstructs rather than repairs: damage across eyes, mouths, and other identifying features.
- Cannot help: a piece of the print that is physically missing with no reference, or a face that was never captured.
Why Is This an App and Not a Web Tool?
Inpainting runs a generative model that needs more memory than a browser tab can reliably provide, particularly at the resolutions people scan prints at. The Evoke AI app runs it on your device using proper hardware acceleration.
How to Repair a Damaged Photo
- Install the Evoke AI app from Google Play or the App Store.
- Open Descratch & Repair from the tools list.
- Load a scan or photograph of the damaged print.
- Let the model detect and fill the damage.
- Check the faces closely at full zoom before saving.
Scan Quality Determines Everything
The most common reason repair results disappoint is a poor capture of the original print. If you photograph a glossy print under a ceiling light, the glare is baked into the image, and the model will treat that glare as part of the photo rather than as damage.
- Use a flatbed scanner where possible. It gives even lighting and no perspective distortion.
- If using a phone, shoot in indirect daylight with the print flat and the camera parallel to it.
- Scan at a high enough resolution that the scratch is several pixels wide — the model needs to see it clearly.
- Do not sharpen or auto-enhance before repair. Those filters make scratches crisper and harder to remove.
- Repair first, then colorize or enhance. Doing it the other way round bakes the damage in.
A Note on Irreplaceable Photos
Always keep the original scan untouched. Save the repaired version as a new file. Automatic repair is very good, but it makes judgement calls, and you may later prefer a different balance between removing damage and preserving the character of an old print. If the photo genuinely matters, a human restorer working manually will still beat any automatic tool on severe damage.
Frequently Asked Questions
No. Descratch & Repair runs only in the Evoke AI mobile app for Android and iOS. The inpainting model needs more memory than a browser tab provides.
Yes. It is free in the Evoke AI app and there is no watermark on the restored photo you save.
It will fill the tear and produce a complete face, but where the damage destroyed the features the result is the model's reconstruction rather than the real person. Inspect faces closely before you treat the result as accurate.
Repair first. Colorizing a damaged photo tends to paint the scratches as if they were real features, which makes them harder to remove afterwards.
Some grain is smoothed along with genuine noise. If the grain is part of the character you want to keep, compare the result against your original before saving.
It targets physical damage such as scratches and creases rather than colour shift. For fading, the enhancer and colorizer are better suited.