🖼️

Restore old photos with AI: deblur, colourize and upscale

Upload a faded, blurry or scratched old photo; let AI sharpen it, fix scratches and colourize black-and-white shots.

Image & Design Beginner

The old photos you dig out at home almost always look the same: yellowed, blurry, creased and scratched — and on black-and-white ones you can barely make out the faces. When you want to build an album, send them to elders, or print and frame them, something’s always missing. Paying someone per photo costs money and takes time.

Now you can do it yourself with AI: upload an old photo and have it deblur, upscale, repair scratches, and even colourize black-and-white shots automatically. The trick: fix one goal at a time, step by step — sharpen first, then colourize, then enlarge — which is more controllable than a single “just fix it” and keeps the result closer to the original photo.

When to use it

Sorting a family album, making a birthday or memorial book for elders, printing old photos to frame, or sharing a blurry old group shot in the family chat — AI can rescue them.

How to do it

  1. Photograph or scan the old photo cleanly — keep it flat, evenly lit, no glare
  2. Open WHEE or Jimeng and upload it via the restore / image-to-image / editing feature
  3. Ask for one thing first — “sharpen and upscale” — then check the result
  4. Once happy, add “colourize the black-and-white naturally / repair this scratch / enlarge to higher resolution”, one step at a time

Weak vs strong

❌ How most people write it
Fix this old photo for me.
✅ Do this instead
Restore this old photo: remove blur and noise, sharpen it to high definition, repair the scratches and creases on faces and background, keep the people’s features and original expressions unchanged, no over-smoothing and don’t alter their looks.

The left says nothing about what to fix or “keep the faces”, so AI may quietly change them; the right pins down deblur, scratch repair, keeping features and no over-smoothing, so it comes out sharp and still looks like them.

Copy-paste prompt

Restore this old photo. Goal:【deblur to HD / colourize the black-and-white / repair scratches and creases / upscale resolution】. Requirements: natural and realistic, keep the people’s features, expressions and composition unchanged, no over-smoothing, don’t alter their looks, colours【natural and undistorted】.

Worked examples

Example 1 · Blurry old photo to HD
Restore this blurry old photo to high definition: remove blur and noise, enhance detail and clarity, repair scratches and creases, keep the people’s features and expressions unchanged, natural and realistic, no over-sharpening.

You get:A noticeably clearer, cleaner photo — faces and details are legible, presentable to send to elders or to print.

Example 2 · Colourize a black-and-white photo
Colourize this black-and-white old photo naturally: keep skin, clothing and background colours realistic and period-appropriate, keep the features and composition unchanged, not over-saturated or unnatural.

You get:The black-and-white shot becomes a natural colour photo and the people suddenly come alive — perfect for a memorial album.

Level up

  • Step-by-step is safest: “sharpen” first, then “colourize”, then “upscale” — one goal at a time beats asking for everything at once
  • Fix locally: mark a single scratch or area and say “only fix here, leave the rest”, preserving the original texture
  • Bring it to life: after restoring, run it through an image-to-video tool so the person smiles or blinks gently — a moving keepsake

Common mistakes

  • Asking for everything in one go — deblur + colourize + upscale + scratch repair at once gets messy and changes faces; split into steps
  • Over-restoring until it’s not them — heavy smoothing and invented detail cause “face swap”; add “keep features and expression unchanged” and compare to the original
  • A bad source shot — glare, tilt or darkness gives AI nothing to work from; clean up the re-shot/scan first

FAQ

Will the restored face stop looking like the real person?
It can, especially when the source is very blurry and AI has to “guess” detail. To reduce the risk: stress “keep the features and expressions unchanged” in the prompt, avoid heavy beautifying/sharpening, and compare against the original to pick the closest version. When a photo is badly degraded, AI inevitably fills in some detail — keep that in mind.
Will the colourized colours be accurate?
Not necessarily. AI infers colours from common sense and period feel; it can’t know the actual colour of a specific garment or wall, so it may differ from reality. If you remember the true colours, just tell it “this coat was navy, the wall was beige” for a closer result.

Pro tip:Save the restored HD version as a new file — never overwrite your original scan. The original negative/scan is always the most valuable; the restored copy is just the “nice and usable” version.

Related tips