Repair a torn or creased photo
A repaired copy of a photo with fold lines, cracks and torn corners filled in. Upload a clear photo, paste this prompt into ChatGPT, Gemini or Grok, change the parts in brackets to suit you, and generate.
Part of the Old photo restoration trend
- Written for
- ChatGPT, Gemini, Grok
- You'll need
- A clear photo to upload
- Status
- Tested 19 Oct 2026 in grok-imagine-image-2.0
- Test settings
- Reference photo: an AI-generated old photo of fictional adults, made to look damaged for this test. Uploaded with the default copy text unchanged. Aspect ratio matched the reference photo (3:2). First result used; resized to 1600x1067 and converted to WebP.
- Variability
- Repairs and colours are the model's guess and vary between runs; always compare with your original photo.
Use the uploaded old photo as the reference and restore it; everyone in it is my family or has given permission. Keep every face exactly as it is: same features, expression, age, skin tone and proportions; do not beautify, slim or reshape anyone; check the result against the original. Scene: the same photo and framing, with deep fold lines and a torn bottom-left corner repaired so the lines no longer show. Pose: keep every pose exactly as in the original. Clothing: keep the original clothes; rebuild only the areas crossed by the damage. Light: keep the original lighting and tones. Style: careful photo repair with missing corners filled with plain background only, matching the original grain and softness. Format: same aspect ratio as the original photo Avoid: changing any face or age, adding or removing people, inventing new objects, over-sharpening, text, logos, watermarks.
Customise
For example: deep fold lines and a torn bottom-left corner.
For example: missing corners filled with plain background only.
Fields start with illustrative values; replace them with your own facts. The prompt updates as you type. Copy prompt copies your version.
How to use it
- Open an external image tool listed above. The tested model, if any, is named in Status.
- Upload a clear reference photo of the subject. If it includes identifiable people, get their consent and check each face.
- Customise the fields, copy your version, paste it with the photo and generate.
- Compare the result with the reference. If a face, pose or detail changes, ask for one correction or try again; an exact match is not guaranteed.
How to use this prompt
- This prompt changes your own photo. Choose a clear reference photo, and get permission from people who can be identified.
- Upload the photo in the external AI app you chose. Fill Customise, press Copy prompt and paste the prompt with the photo.
- Compare the new picture with the original. Face, hands, clothes and pose can change. An exact likeness is not guaranteed.
Good to know
- Where a crease crosses a face, the repair is the model's guess; check that face most carefully.
- Use reference photos you have the right to use; get consent from identifiable people.




