aionDownload Aion

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.

Related guides

How to

Step by step

  1. Upload the best scan you have

    A sharp, evenly lit, full-resolution scan. Crop out borders and album edges.

  2. Ask for a faithful restoration

    Say what to fix: scratches, fading, blur. Say what to keep: faces, expressions, background, era.

  3. Ask it not to invent detail

    State explicitly that missing areas should be left plausible and faces must not be changed.

  4. Generate a few versions

    Outputs vary. Compare several rather than trusting the first.

  5. Check faces against the original

    Zoom in. Look for changed features, smoothed skin, altered eyes or teeth.

  6. Reject anything that changes a likeness

    If the person no longer looks like themselves, the result is a new image, not a restoration.

Questions

Common questions

Is ChatGPT or Gemini good at restoring photos?

They are quick and handle mild damage and fading well. Their weakness is fidelity. Because the underlying models generate images, they can subtly or badly redraw faces and details.

Why did the face change after restoration?

General image models reconstruct the whole image, including faces, from their training rather than repairing the pixels that are there. A dedicated restoration model is constrained to preserve existing features.

Is it free?

The free tiers of these assistants allow limited image work. Heavier use needs a paid plan. Dedicated restoration apps set their own pricing, shown on their store listing.

Can I get a print-resolution file out?

Often the output is limited in resolution. You may need to upscale it separately for print.

When should I use a dedicated tool instead?

When the photo matters, when faces must stay exact, or when you want a repeatable process. A restoration-specific tool trades flexibility for fidelity.