Restore an old photo with AI

Repair the scratches, stains and creases of an old print and colorize it with one instruction, then generate the restoration video where the photo gradually comes back to life.

Interactive demo

Real workflow, real results. Pan and zoom freely.

Use this workflow

The real workflow: the damaged 1960s print is restored and colorized by instruction, and the video replays the transformation from one to the other.

At a glance

What the workflow produces: 1 image in 1:1 · 1 video of 5s in 1080p, 1:1
Canvas structure: 5 nodes wired by 5 connections
Models wired in: Flux Kontext, Kling V3
Cost of one full run: about 158 credits, roughly €1.58

Family photos age badly: fading sepia, scratches, creases, dog-eared corners. This workflow repairs everything in one step: the photo's scan (Media node) goes through an Image node driven by a restoration instruction that erases the damage, colorizes with credible period colors, and above all keeps the faces exactly as they are.

That is what separates a restoration from a reinterpretation: Flux Kontext is an editing model, it does not reinvent your grandparents, it cleans the print. The instruction explicitly locks faces, pose and framing; only the damage and the missing color are treated.

The restoration video is the moving extra: before and after frame 5 seconds where scratches fade and colors emerge, generated by Kling V3. A birthday gift that lands, for less than a euro fifty total.

How it works

1
Scan or photograph the print
A flatbed scan or a well-lit photo without reflections. Import the file into the workflow's Media node.
2
Adjust the restoration instruction
The default instruction erases scratches and stains, repairs creases and colorizes. Want to stay black and white? Simply remove the colorization part.
3
Generate the restored photo
The Image node produces the repaired print, same faces, same framing. Regenerate with a refined instruction if a detail resists.
4
Create the restoration video
Before and after become the first and last frames: the photo repairs itself on screen, a perfect format to gift or share.

Variations and use cases

The same canvas, tuned for different needs: each variation is two or three lines of prompt away.

Repairing a torn or creased photo

Scratches, folds, missing corners: the editing instruction erases them while keeping faces intact. Flux Kontext repairs rather than reinvents, which is the whole difference on a family photo.

Colourising a black and white photo

Colourisation alone is a separate instruction: drop the repair part. Ask for period-accurate colours rather than saturated ones, that is what avoids the postcard look.

Grandparents' wedding photo

The most common case, and the most sensitive: check faces pixel by pixel before gifting the print. Always keep the original next to it on the canvas, comparison is the only check that counts.

Making an ancestor's photo move

The restoration video replays the transformation in five seconds: scratches fade, colours emerge. A gift that lands, for about 1.50 euros with Kling V3.

Yellowed, stained or overexposed prints

Sepia shift and damp stains are handled in the same instruction. A completely burnt-out area, on the other hand, holds no information left: the model will invent, and it will show.

Common mistakes

A scan shot at an angle or with glare

Glare is read as a bright area of the photo and carries into the render. Flatbed scan, or a daylight photo without flash.

Not locking the faces

Without an explicit instruction, an editing model adjusts features and improves smiles. The photo becomes unrecognisable to those who knew the person.

Restoring and cropping in one pass

Two intentions in one instruction give an unstable result. Restore first, crop afterwards in a second node.

Recommended models

Flux Kontext

Kling V3

Nano Banana

Frequently asked questions

Are the faces really preserved?

Yes, the instruction locks them explicitly and Flux Kontext is an editing model, not free generation. Always compare with the original: that is the advantage of having both side by side on the canvas.

Can I colorize only, without repairing?

Yes, the instruction is freely rewritable: colorization only, repair only, or both. Each variant is a generation costing a few cents.

What scan quality do I need?

The sharper the scan, the better the result. A well-lit smartphone photo without reflections already works very well.

Can I have the restored photo printed?

Yes, the definition covers a framed print up to A3. For a very large format, run the file through an upscaler before printing.

What about an area that is completely torn away?

The model rebuilds it by inventing. On a background or a garment that goes unnoticed, on a face you should reject the render: no restoration guesses features that are gone.

Can I restore a whole album?

Yes, one Media node and one Image node per photo, with the same instruction wired into all of them. Twenty prints come to 1.20 euros of credits.

Can a small old photo be enlarged?

Editing improves the render but does not create missing detail: a six-centimetre print stays limited. Scan at 600 dpi rather than photographing it up close.

Restore an old photo with AI

Use this workflow

Free account, no card required. The workflow opens pre-filled in your canvas.