Photo Restoration Triage: Which Photos Are Worth Saving and Which Are Beyond Repair

You found a shoebox of old family photos in your parents' attic. Some are in surprisingly good shape — a little faded, a little yellowed, but the faces are clear. Others look like they went through a washing machine: torn corners, water damage, a mysterious brown stain that might be coffee from 1978. You want to restore them all, but an AI photo restorer costs credits and time. Some photos will restore beautifully. Some will never look good no matter what you do. Here is how to triage before you start.

The triage framework: green, yellow, red

Sort your photos into three piles before you run a single one through the restorer:

🟢 Green — will restore well (start here): Fading, yellowing, low contrast, minor dust and scratches, slight blur. These are the easiest problems for AI to fix. Color fading gets corrected by histogram equalization. Yellowing gets white-balanced. Dust and scratches get inpainted using surrounding pixel data. Expect 80-95% improvement with one pass through the photo restoration tool.

🟡 Yellow — might restore with multiple passes: Moderate tears (especially through non-face areas), creases that do not cross eyes or mouths, larger stains on backgrounds, photos that are both faded AND scratched. These need the restorer, then manual touch-up, possibly a second restoration pass. Budget 2-3 credits per photo. Results will be noticeably better but not perfect.

🔴 Red — probably not worth it: Missing chunks (torn-off corners with actual image data gone), damage directly over the center of a face (the AI will guess, and guessed faces look wrong), extreme water damage where the emulsion has separated from the paper, photos that are more damage than photo. The AI cannot restore what is not there. If the original detail is gone, the AI invents plausible detail — which might look fine at a glance but wrong to anyone who knew the person.

The specific damages and what AI can do about each

Fading and color shift → Excellent fix. This is what AI restoration does best. The model has seen millions of properly-exposed photos and knows what natural skin tones, sky colors, and vegetation should look like. Faded photos come back to life with realistic color and contrast. Pair with the AI colorizer if the original is black and white.

Scratches and dust → Very good fix, with caveats. Thin scratches across backgrounds or clothing get filled cleanly. Scratches across faces are harder — the AI fills the scratch but may slightly alter the facial feature. A scratch through an eye might result in an eye that looks slightly different from the other one. Always zoom in and check facial features after restoration.

Tears and missing pieces → Partial fix. If the tear is through a uniform area (sky, wall, grass), the AI fills it seamlessly. If the tear is through a detailed area (face, text, patterned clothing), the fill will be approximate. For torn corners with sky or simple background, the AI reconstruction is usually good enough. For a tear that removed half of someone's face, no AI can reconstruct what that person actually looked like.

Water damage and mold → Unpredictable. Water damage creates irregular patterns that confuse the AI — it is not sure what is damage and what is actual image content. Mold spots are often treated as image features rather than defects. These photos need manual pre-cleaning (digitally clone out the worst spots) before AI restoration.

The restoration pipeline order

If a photo needs multiple fixes, the order matters: restore first (fix the damage), then colorize (add color if black and white), then upscale last (increase resolution). Restoring after upscaling means you are fixing damage at higher resolution, which is slower and produces worse results. Colorizing before restoring means the colorizer works on damaged pixels, producing color artifacts that the restorer then has to fix. Always restore first.

Start with your green-pile photos. Run 5-10 through the photo restorer. See what the AI can do. Then decide whether your yellow-pile photos are worth the extra effort. The red pile — keep the originals, but do not spend credits on them. For the complete photo enhancement workflow, see our guide to the correct order of operations for photo restoration.