How to Restore Old Photos With ChatGPT or Gemini
General assistants like ChatGPT and Gemini can restore old photos from an uploaded scan and a prompt. They are convenient but use image models built to generate, not to repair, so they can change faces, backgrounds, and details. Always compare the output with the original and reject results that alter a likeness.
They can, with a caveat
Upload a scan to ChatGPT or Gemini, ask for a restoration, and you will get a cleaned-up image in seconds. For a lightly faded or scratched photo, the result can be good. The caveat is what these tools are: general image models built to generate new pictures, adapted to restoration. Generation is exactly the risk.
Write the prompt defensively
Name the damage to fix and, more importantly, name what must not change: the faces, the expressions, the background, the period look. Add that missing areas should be left plausible rather than invented. This reduces, but does not remove, the chance of a redrawn face.
Generate several, trust none by default
Outputs vary between runs. Make a few and compare. Then do the one check that matters: zoom into every face and compare it with the original. Look for shifted features, over-smooth skin, changed eyes, new teeth.
The line between restoration and replacement
If the person in the output no longer quite looks like the person in the original, you do not have a restored photo. You have a new image based on it. For a stranger in the background that may be fine. For your grandmother it is not.
When to use a dedicated tool
A restoration-specific model is constrained to preserve what is in the frame. You give up the open-ended flexibility of a chat assistant and get fidelity and a repeatable process in return. That trade is worth it when the photo is one you cannot replace.
Doing it on iPhone with Aion
Aion uses a separate model per job, each built to repair rather than regenerate, with a before and after view on every result. See how to restore old photos for the full method and how much photo restoration costs for the options compared.