Use case · AI Photo Restoration
Restore an old family photo with AI
Old prints fade, crack, and yellow. Newer phone photos blur and lose detail. ImageSuite\'s AI Photo Restoration runs two different models — one for face recovery, one for physical-damage repair — and lets you pick which one matches your photo. Built on CodeFormer and Microsoft\'s "Bringing Old Photos Back to Life", both via Replicate.
Which mode for which damage
Face mode — CodeFormer
Best when the photo is mostly intact but faces are blurry, low-resolution, or have lost sharpness. CodeFormer reconstructs facial detail from a learned prior of millions of portraits.
- · Blurry headshots from older phones
- · Soft-focus portrait prints
- · Low-resolution scanned faces
- · Group photos where individuals look fuzzy
Damage mode — Microsoft
Best for physical damage to the print itself — tears, scratches, creases, water spots, sepia fading. Trained on a dataset of artificially damaged historical photos.
- · Folded or creased prints
- · Scratches across faces or backgrounds
- · Faded sepia or yellow-tinted images
- · Water-spot or stain damage
When in doubt, try damage mode first. If the faces still look soft, run the result through face mode in a second pass.
Restore your photo now
Pro feature — $9/month flat. No per-image credits.
How to scan an old print well
The restoration result is only as good as the input. A few minutes spent on the scan or capture step pays back many times over.
- 1. Flat-bed scanner beats phone camera. A 300+ DPI flat-bed scan keeps the photo flat and lit evenly. If you only have a phone, place the print on a flat dark surface and shoot from directly above with diffuse natural light (no flash).
- 2. Crop minimally. Leave some white border around the photo edge in the scan — the AI handles cropping later. Hard-cropping the print can chop off subtly damaged edges that hold useful structural information.
- 3. Skip pre-processing. Don\'t try to fix the photo in Photoshop first. Both models perform better on the unmodified scan.
- 4. Save as PNG. JPEG\'s lossy compression bakes in artifacts that the restoration model then tries to interpret. PNG keeps the scan honest.
What to expect — realistic outcomes
Usually restores well
- · Light to moderate scratches
- · Sepia and yellow fading
- · Soft / out-of-focus portraits
- · Minor water staining
- · Mild creasing along folds
Harder cases
- · Large missing chunks of a face
- · Severe chemical damage / mold
- · Heavy double-exposure or stains over faces
- · Photos already heavily edited / re-touched
For the hard cases, AI restoration is a strong starting point but expect to combine with manual touch-up in a photo editor for the best result. ImageSuite\'s Photo Editor can clean up small residual artifacts.
FAQs
How do I restore an old family photo digitally?
Scan the photo flat (or take a well-lit, square-on photo of it with a phone), upload it to ImageSuite's AI Photo Restoration tool, pick face mode for portraits or damage mode for tears and scratches, then download the restored version. CodeFormer handles the face detail; Microsoft's "Bringing Old Photos Back to Life" handles physical damage.
Will AI restoration change what the people look like?
Some change is unavoidable — the AI is reconstructing detail that isn't in the source. CodeFormer is unusually faithful for face restoration but may slightly soften or shift expression. For irreplaceable family photos, always keep the unrestored original alongside the restored version.
What kind of photo damage can be fixed?
Damage mode handles scratches, creases, tears, water stains, sepia fading, and uniform colour shifts. It struggles with large missing areas (where most of a face or important object is gone) and heavy chemical damage that has obliterated detail. Light to moderate damage typically restores well.
Can I colorize the restored photo too?
Yes — run the restored output through the AI Colorize tool (DDColor). The recommended order is restore first (to recover detail), then colorize (to add colour to the cleaner image). Colorizing before restoring tends to produce worse colour because the model is matching damaged tones.