How to restore and colorize old photos with AI (2026)
Quick answer
To restore and colorize an old photo, digitize the best original you can find, keep that scan untouched, then use Old Photo Restoration when the image has damage or Photo Colorizer when it mainly needs color. Compare faces, lettering, uniforms, buildings and other historically meaningful details before sharing the result.
AI restoration reconstructs plausible detail and AI colorization predicts colors. Neither process proves what missing details or the original colors actually were.
Choose the right old-photo workflow
Workflow | Use it when | What it changes | Historical limit |
|---|---|---|---|
The scan has scratches, stains, fading, softness or lost contrast | Repairs visible damage, improves clarity and creates a color version | Missing detail and original colors are inferred | |
The photo is structurally usable and mainly needs color | Adds an automatically predicted color version | Current tool has no manual color fine-tuning |
Restore one old photo
Upload a copy of the highest-quality scan you have. Compare faces, lettering, clothing, architecture and damaged areas with the untouched original before downloading or sharing the result.
Open Old Photo Restoration1. Create the best digital source you can
Use a flatbed scan when available. If the photo can only be photographed, place it flat, use even indirect light, keep the camera parallel to the print and avoid reflections. Save the highest-quality original before cropping, compressing or editing it.
Clean the scanner glass or camera lens.
Capture the whole print, including edges.
Avoid digital zoom and social-media copies.
Save an untouched archival file and work from a duplicate.
2. Decide whether the photo needs repair, color or both
Use Old Photo Restoration when scratches, stains, folds, fading, softness or lost contrast need repair. The current Magic Hour page says this workflow also creates a natural-looking color version, so a second colorization step is not automatically required.
Use Photo Colorizer when the scan is already structurally usable and the main goal is an automatically colorized version. The current colorizer does not offer manual color fine-tuning, so treat clothing, signs, skin tones, vehicles and period-specific colors as estimates.
3. Generate from a duplicate
Open the selected tool, upload a JPEG or PNG copy and generate one result. Check the live page for the current free allowance and download terms rather than relying on an old credit or timing figure.
4. Compare the result with the original
Faces: identity, eye shape, teeth, hairline and expression.
Text: names, dates, signs, uniforms and insignia.
Damage: whether a repaired tear or stain caused a new object or repeated texture.
Geometry: hands, jewelry, furniture, buildings and straight edges.
Color: skin, clothing, vehicles, flags and materials that may have historically specific colors.
If the result changes an important fact, keep the original area or repair it manually. A visually convincing reconstruction is still an inference.
5. Upscale only after approving the restoration
If the accepted image is too small for print, use AI Image Upscaler after the restoration and color choice is settled. Compare the upscale with both the accepted restoration and the original scan; upscaling can create plausible fine texture rather than recover documented detail.
How to preserve provenance
Keep three files: untouched scan, restored result and colorized or upscaled derivative.
Record the tool and date. Note which workflow produced each derivative.
Label inferred color. For archives, journalism or education, say that color was generated rather than documented.
Retain source evidence. Keep inscriptions, envelopes, dates and family notes with the original file.
Animate a restored portrait only as a creative derivative
Talking Photo can generate motion from a restored still and supplied audio. That motion is newly synthesized; it is not recovered footage. If you use it for a memorial, documentary or historical presentation, label the animation clearly and keep the still image available. See the talking-photo tutorial for the separate workflow.
For an actual black-and-white recording, use Video Colorizer. Colorizing recorded frames and inventing motion from a still image are different operations.
Frequently asked questions
Use the restoration workflow when the source has damage, fading, noise or softness. Magic Hour's current Old Photo Restoration already creates a color version, so use the separate Photo Colorizer when the image is structurally usable and color is the main task.
No. It can reconstruct a plausible face, fabric or background, but a model cannot verify detail that is absent from the scan. Compare with other photographs and retain the original whenever historical accuracy matters.
They may look plausible, but the model does not know the original garment, paint, vehicle or sign colors from grayscale alone. Use documented references when exact color matters and label automated colorization as inferred.
The current Magic Hour restoration and colorizer pages offer limited browser use without sign-up. Check the live tool for the current allowance, output resolution and commercial-use terms before a large or commercial project.
Check the current plan terms and your rights to the source photo, people and publication context. A tool's output permission does not grant rights you did not have in the original image.
