How to Remove Scratches from Old Photos with AI

Why old photos deteriorate and what AI can fix

Understanding how photos deteriorate helps you set realistic expectations for AI restoration. Surface scratches — the most common type of damage — happen when prints rub against each other in a stack, get dragged across a table, or are handled with rough fingers. These scratches remove or displace the emulsion layer on the print surface but do not destroy the underlying image information. AI excels at fixing these because the surrounding context is fully intact, and the scratch is essentially a thin line of missing data that the AI can interpolate from both sides.

Creases and folds create broader damage zones where the emulsion has cracked and sometimes separated from the paper base. The photo still contains the image information on either side of the crease, but the fold line itself shows as a white or discolored stripe, sometimes with slight misalignment where the paper was bent. AI handles creases well because it can reference the intact areas on both sides to reconstruct the damaged strip, though deep creases across high-detail areas like faces may require multiple passes.

Tears and missing sections present the greatest challenge. When part of a photo is physically torn away, the image information is gone — the AI must generate plausible content based entirely on context. A torn corner showing sky and trees is easy to reconstruct because the AI can extend the existing patterns. A tear through a face is harder because the AI must infer facial features from whatever remains visible. Modern AI tools handle this surprisingly well for moderate damage, though severe facial damage may still benefit from professional restoration.

• Surface scratches are the easiest to fix — the AI interpolates from intact pixels on both sides of the scratch.
• Creases and folds leave broader damage strips but usually preserve context on both sides for accurate reconstruction.
• Tears with missing sections require the AI to generate new content based on surrounding context.
• Water stains and chemical discoloration respond well to AI color correction after the surface damage is repaired.

Scanning: the foundation of a good restoration

The quality of your scan determines the ceiling for your restoration. A low-resolution phone snapshot of a scratched photo gives the AI very little information to work with — the scratches blend into the image noise, and there is not enough surrounding detail for accurate reconstruction. A proper 600 DPI flatbed scan captures the full detail of both the image and its damage, giving the AI a clear map of what needs to be fixed and plenty of context to draw from.

Scan in color even if the original photo is black and white. Color scans capture subtle tonal variations in the paper, the emulsion, and the damage that a grayscale scan flattens out. These tonal differences help the AI distinguish between intentional image content and damage artifacts. You can always convert to grayscale after restoration if you prefer a black-and-white final image.

Do not attempt to clean the photo aggressively before scanning. Gently remove loose dust with a soft brush, but do not try to wipe away stains or flatten creases by force — you risk causing additional damage. Scan the photo in its current state and let the AI handle the digital cleanup. The one exception is loose fragments: if a tear has created a separated piece, position it as accurately as possible on the scanner bed before scanning so the AI has both pieces in the correct spatial relationship.

Removing scratches and creases step by step

Start with the simplest damage and work toward the most complex. Begin with light scratches on uniform backgrounds — sky areas, walls, clothing with even texture. These give you a feel for how the Magic Eraser brush works and build confidence before tackling scratches across critical areas like faces. Select a brush size slightly wider than the scratch so you cover the full width of the damage in a single stroke.

For scratches that cross multiple textures, break the work into segments. A scratch running from a blue sky through brown hair and across a white shirt crosses three distinct texture zones. If you trace the entire scratch in one stroke, the AI tries to reconcile all three textures simultaneously and may produce artifacts at the transitions. Instead, process the sky segment, then the hair segment, then the shirt segment independently. Each segment gives the AI a consistent texture context to work within.

Deep creases that have caused the emulsion to crack may require two passes. The first pass reconstructs the major damage — the white or discolored stripe of the crease itself. The second pass cleans up any remaining artifacts or slight color mismatches along the edges of the repair. Zoom to 100% or 200% after each repair to verify the result before moving on. It is much easier to redo a single scratch repair than to figure out which repairs need rework after processing the entire photo.

• Start with simple scratches on uniform backgrounds before tackling complex areas.
• Break long scratches into segments where they cross different textures for cleaner results.
• Use a brush slightly wider than the scratch to cover the full damage width in one stroke.
• Deep creases may need two passes — one for major reconstruction, one for edge cleanup.

Restoring faded color and contrast

Once all physical damage is repaired, the photo likely still looks aged — faded colors, low contrast, and a yellow or magenta color cast from decades of chemical deterioration. AI Enhance addresses these issues in a single pass. The tool analyzes the overall color balance and corrects the shift, restores contrast between light and dark areas, and sharpens details that have softened with age.

Color restoration is particularly dramatic for photos from the 1960s through 1980s that were printed on consumer-grade paper. These prints fade unevenly — reds and yellows deteriorate faster than blues and greens, which is why old photos often develop a blue-green cast as the warm tones fade. AI Enhance recognizes these chemical aging patterns and restores the color balance to approximate the original print tones. Skin tones, which are especially sensitive to color shifts, typically improve significantly.

Be conservative with enhancement settings if the photo will be compared to the original print. Slight remaining warmth or softness can feel more authentic than a perfectly corrected image that looks like it was taken with a modern camera. The goal is restoration — bringing the photo back to how it looked when it was printed — not modernization. Save both the restored version and the enhanced version so you can choose which feels right for the context.

Preserving and sharing your restored photos

Save the fully restored image in a lossless format — TIFF or PNG — at the full scan resolution. This is your archival master copy. Every future use of the image should be derived from this master rather than re-editing the original scan. Store the master file in at least two locations: a local hard drive and a cloud storage service. Physical media fails, cloud services change terms, and redundancy is the only reliable protection for irreplaceable files.

For sharing, export smaller versions in JPEG or WebP format. A 2000-pixel-wide export at quality 85 produces a file small enough to email or upload to social media while retaining excellent visual quality. If you plan to print the restored photo, export at the full resolution and send the file to a photo printing service that accepts TIFF or high-quality JPEG. Modern printing services can produce remarkable results from well-restored scans — the print will look better than the original because the damage is gone.

Consider organizing your restored photos into a digital family archive. Add metadata — names, dates, locations, relationships — to each file so future generations can identify the people and places in the photos. The combination of AI restoration and careful metadata creates a family archive that is more accessible, more durable, and more complete than the original shoebox of prints.

• Save archival masters in lossless TIFF or PNG at the full scan resolution.
• Store master files in at least two locations — local storage and cloud backup.
• Export sharing versions at 2000 pixels wide in JPEG or WebP for email and social media.
• Add metadata with names, dates, and locations to build a searchable digital family archive.