
How to Remove Scratches from Old Photos: AI + Manual Cleanup
A practical workflow for scanning scratched photos, using AI scratch repair, and finishing difficult areas with manual cleanup without inventing too much detail.
ArtImageHub Team
Scratches are one of the most common old-photo problems, but they are not all the same. A thin surface scratch across a plain background is easy to repair. A deep gouge through an eye, mouth, or handwritten note is a different problem.
The best workflow is usually AI first, manual cleanup second. Let AI remove the many small defects quickly, then spend human attention only where it matters.
Do not physically fill scratches before scanning
Old home advice sometimes suggests petroleum jelly, oil, varnish, or other coatings to make scratches less visible. Do not do that before digital restoration.
Those treatments can:
- leave residue on the print,
- attract dust,
- affect scanner lighting,
- smear softened emulsion,
- make future conservation harder,
- create artifacts the AI has to work around.
Instead, remove loose dust gently with a clean blower or soft brush if the print surface is stable. Then scan the photo untreated.
Scan settings for scratched photos
A good scan gives the repair model context around each scratch.
Use:
- 600 DPI for most family prints.
- 1200 DPI only for very small originals or highly detailed portraits.
- color mode, even for black-and-white prints.
- PNG, TIFF, or high-quality JPEG.
- scanner glass cleaned before scanning.
- no aggressive auto-sharpening.
If the scanner has a scratch-removal feature, be cautious. Some scanner corrections soften detail or create interpolation artifacts. For AI repair, a clean honest scan is usually better than a pre-smoothed one.
What AI handles well
AI scratch repair works best on scratches that are narrow enough for surrounding context to guide the fill.
Good candidates:
- fine white scratches,
- dark scuffs,
- dust lines from scanning,
- small cracks in plain backgrounds,
- scratches across clothing,
- surface marks that do not remove large areas.
ArtImageHub is our own browser-based restoration workflow. Use ArtImageHub when you want a one-time $4.99 AI pass to remove scratches from a scanned old photo, then compare the restored version against the untouched scan.
What AI cannot know
When a scratch removes image material, AI is not "recovering" the exact original. It is synthesizing a plausible fill from nearby information.
This distinction matters when scratches cross:
- eyes,
- mouths,
- hands,
- uniforms,
- medals,
- documents,
- signs,
- handwritten captions.
For family display, a plausible repair may be exactly what you want. For historical documentation, keep the original scan and note that the restored version includes AI reconstruction.
The AI + manual workflow
Use this order:
- Scan the original photo.
- Save the untouched scan.
- Run AI scratch repair.
- Compare before and after at normal viewing size.
- Mark only the remaining distracting defects.
- Use manual heal/clone tools for those areas.
- Export restored full-resolution and sharing copies.
Do not spend an hour manually fixing every scratch before AI. The AI pass may change those areas again. Do broad automated repair first, then local human cleanup.
Manual cleanup tips
For remaining scratches, use small strokes.
Practical rules:
- sample from nearby texture,
- follow the direction of clothing or background,
- avoid copying one eye or mouth detail onto another,
- zoom out often,
- preserve some natural grain,
- stop before the photo looks airbrushed.
For faces, work more slowly. A scratch through a cheek is easier than a scratch through an eye. If the repair changes expression, undo and use a more restrained edit.
Common scratch types
Hairline surface scratches:
- Usually good AI candidates.
- Often disappear with little manual work.
Parallel scanner scratches:
- Check scanner glass first.
- Rescan if the line came from dust or glass, not the original print.
Cracked emulsion:
- AI can reduce the visual break.
- Some texture may remain.
Deep gouges:
- AI may fill the gap plausibly.
- Important details may still need manual correction or a softer crop.
Network scratches:
- Many crossing scratches are harder.
- Repair may work better in stages: AI pass, review, then manual cleanup of the worst intersections.
When to leave a scratch
Not every scratch needs to disappear.
Leave or soften a mark when:
- removing it changes the person's face,
- the scratch is part of the print's history,
- repair creates obvious invented texture,
- the mark is invisible at normal viewing size,
- the photo is being used as evidence or archive material.
Old photos do not need to look new. They need to be readable and emotionally true.
Final review
Before sharing or printing:
- compare to the untouched scan,
- check faces first,
- check hands and text second,
- view at normal size,
- print a small proof for important photos,
- keep the original scan with the restored file.
The best scratch repair is the one nobody notices. If the viewer sees the person before they see the restoration, the workflow did its job.
About the Author
ArtImageHub Team
Photo Restoration Editors
The ArtImageHub team writes practical guides for restoring, preserving, and sharing old family photos with AI and careful manual workflows.
Share this article
Ready to Restore Your Old Photos?
Try ArtImageHub's AI-powered photo restoration. Bring faded, damaged family photos back to life in seconds.
