---
title: "Restore an old or damaged photograph"
description: "Restore photo repairs scratches, tears and faded colour with FLUX Kontext. What it recovers faithfully, what it reconstructs, and why that distinction matters."
canonical_url: "https://justpictur.ing/guides/restore-an-old-photo/"
published: 2026-08-14
updated: 2026-08-14
locale: "en"
tools: ["restore-photo"]
---

Restore photo takes a damaged photograph and repairs it: scratches filled,
creases removed, torn areas rebuilt, faded colour brought back toward what it
once was.

No prompt, no settings. One photograph in, one repaired photograph back.

## What "restore" honestly means

There are two different operations behind that one word, and it is worth
knowing which one you are getting on any given part of the picture.

**Repair** is undoing damage that sits _on top of_ the image: a scratch across
the emulsion, a crease, dust, a stain, a colour cast from decades of light. The
information underneath survived, and putting it back is close to objective.

**Reconstruction** is filling in what is _gone_: a missing corner, a face
scratched through, an area torn away. Nothing survived there, so what appears is
plausible rather than recovered. It will look right. It is not a record.

> A restored photograph is a picture of what the photograph looked like. Where
> the damage removed the subject, it becomes a picture of what the subject
> plausibly looked like, and only you know which parts those are.

That distinction is the whole of the judgement this tool asks of you. It matters
most on faces, on writing, and on anything the photograph was kept as evidence
of.

## Getting the best out of it

**Scan or photograph the original as flatly as you can.** Even light, no glare,
the whole print in frame and square to the lens. The tool is repairing what you
give it; a highlight blown out by your phone's flash is damage you added, and
it will be treated as part of the picture.

**Give it the full print, not a crop of the interesting part.** Restoration is
easier with surrounding context, because texture, grain and colour are inferred
from the parts that survived. Crop afterwards.

**Do this before enlarging, not after.** Damage that is repaired at the original
size is repaired once; damage that is enlarged first is four times as much
damage to fix and comes back with the repair enlarged too. Restore, look, then
[Upscale](/guides/upscale-an-image/) if the destination needs the pixels.

**It takes GIF as well as PNG, JPEG and WebP**, the only tool here that does,
which is worth knowing, because an awful lot of family photographs live on old
hard drives in exactly that format.

## What it is not for

- **A modern photograph that is merely soft or noisy.** That is a sharpening job
  ([Upscale](/guides/upscale-an-image/)), and running a restoration over an
  undamaged picture invites the model to change things nothing was wrong with.
- **Colourising a black-and-white photograph.** Fading is a drift away from
  colours that were captured; a monochrome print never had any. What comes back
  would be invented wholesale.
- **Making the picture a better photograph.** It repairs; it does not reframe,
  relight or restage. If you want the subject somewhere else, you want a
  generation with this photograph as a reference.

## When it goes wrong

### A face changed

The most common complaint and the most important one. Faces are where a model's
priors are strongest, so a face that was damaged, tiny or badly out of focus is
the area most likely to be rebuilt rather than repaired, and rebuilt into
something subtly, uncannily not-them.

> **Fix:** compare at full size before you keep it, and keep the original file
> whatever you decide. If the face is the point of the photograph and the face
> was damaged, this is a job for a human retoucher who can be told who that is.

### Something was removed that was meant to be there

A hairline crack, a deliberate scratch, a scuff on a painted surface, a caption
written on the print: all of it reads as damage.

> **Fix:** crop the area out before restoring and put it back afterwards, or
> accept the version that keeps the mark and fix the rest by hand.

### The colours are warmer or cooler than you remember

Correcting a cast means deciding what the neutral was, and the tool infers it
from the picture. Your memory of the day is a different source.

> **Fix:** treat the result as a good base rather than a final grade, and set
> the temperature yourself where it matters.

### The texture went flat

Film grain is noise, and repairing damage means smoothing noise. On a heavily
restored print the grain can go with the scratches, which reads as modern rather
than as old-and-clean.

> **Fix:** this is the trade the tool makes, and it makes it in one direction. If
> the grain is part of what the photograph _is_, restore a copy and blend the two
> yourself.

---

## About justpictur.ing

justpictur.ing is an AI picture studio: prompt-driven images, clips, voice-overs and music, and storyboard-backed films with consistent characters across every scene.

Agents connect over MCP and can generate directly:

```
claude mcp add justpicturing https://api.justpictur.ing/mcp --transport http
```

The app is at https://app.justpictur.ing and requires an account.

Every guide on this site has a Markdown twin at its own URL plus `.md`. Every guide: https://justpictur.ing/guides/