AI Colorize Photo

Add natural colour to black-and-white photographs with a real neural network running in your browser. Free, no upload, no watermark, no sign-up.

100% Free

Upload the photo

…or press Ctrl+V to paste a scan

Drag & drop a black-and-white photo

or

JPG, PNG, WebP — up to 30MB
The colour is predicted on your own device — the photo never leaves it.

Colourise

— the model is about 215MB, downloaded once and then cached by your browser.

The first run downloads the model, which takes as long as your connection needs for 215MB. Colourising itself is about 30 seconds of real compute on your device.

How far the predicted colour is pushed. 100% is what the network predicted; below that is muted, above that is more saturated.

Strength is applied while the colour is painted back in, so changing it runs the model again — roughly another 30 seconds. The result does not update until you press Re-apply.

Before / after

100% Free

All tools are completely free to use.

Privacy First

Your files are never uploaded or stored on our servers.

Super Fast

Compress and process files in seconds.

High Quality

Best results with minimal quality loss.

What is AI Colorize Photo?

AI Colorize Photo brings black-and-white and sepia photographs into colour. Upload a scan, wait about half a minute, and the photograph comes back with skin, sky, foliage, wood, fabric and food coloured the way the scene most likely looked. The network doing the work is DDColor, and it runs inside your browser — the photograph never leaves your device.

Your photo keeps every detail it had

Colour and brightness are separate things, and this tool only touches colour. A photograph is split into lightness and two colour channels; the model predicts only the colour channels, and they are recombined with your original lightness at full resolution. Every grain, highlight, wrinkle and scratch survives exactly as scanned — the result is your photograph wearing colour, not a repainted imitation of it. That is also why a 12-megapixel scan comes back at 12 megapixels even though the model itself works at 512 pixels: colour varies slowly across a scene, so carrying it up to full resolution is invisible, while doing the same to detail would not be.

Predicted, not recovered

This is the honest part, and it matters. A black-and-white negative records no colour at all, so nothing anywhere in the file says whether that jumper was red or blue. The model has learned what the world usually looks like — grass is green, skin is skin-coloured, skies are blue near the top — and applies that. It is convincing on everyday scenes and genuinely uncertain about man-made objects whose colour was arbitrary: cars, clothing, painted walls, signage. Where it guesses wrong, it guesses plausibly, which is the best any colouriser can honestly offer. Treat the result as a beautiful interpretation of the photograph rather than a recovered original.

The model is downloaded, not rented

No public browser-ready build of DDColor existed, so we exported one from the published research weights and serve it from this site as a 215MB file. Your browser downloads it once, caches it, and every later photograph — today or next month — colourises with no download at all. That is why there is no account, no credit system and no upload: the network is running on your machine, so there is nothing for us to meter and nowhere for your family photographs to go.

Nothing is uploaded. Tools on this page that use a neural network download the model to your browser once, cache it, and run it on your own machine from then on. That is why there is no account, no queue, no watermark and no daily limit — there is no server doing the work, so there is nothing for us to meter and no copy of your photo anywhere but your device.

How to Use AI Colorize Photo

  1. 1

    Upload a black-and-white photo

    Drag it in, paste with Ctrl+V or choose a scan. It stays on your device the whole time.

  2. 2

    Press Colourise

    The first run downloads the 215MB model with a real progress bar; after that it is cached. Colourising is about 30 seconds of on-device compute.

  3. 3

    Compare against the original

    Drag the before/after slider. Your original lightness is untouched, so only the colour differs.

  4. 4

    Adjust the colour strength if you want

    Push it up for a bolder result or down toward a gentle tint, then re-apply.

  5. 5

    Download or keep restoring

    Save as PNG, JPG, WebP or AVIF — or hand the photo to the restoration tool for faces, or the object remover for scratches.

Frequently Asked Questions

Predicted. A black-and-white photograph records no colour information whatsoever, so no tool on earth can recover the original hues. The model applies what the world usually looks like — convincing for skin, sky, foliage, wood and food, and genuinely uncertain for things whose colour was arbitrary, like clothing and cars. It guesses plausibly, and this page says so rather than claiming otherwise.
No. Only the two colour channels come from the model; your photograph's original lightness is kept at full resolution and recombined with them. Every grain and detail is preserved, and a 12-megapixel scan comes out at 12 megapixels.
No. The model — about 215MB — is downloaded to your browser instead, once, and cached. Everything after that happens on your own processor. For family photographs, that difference is the whole point.
The first run waits for the 215MB model to download, with a real progress bar. After that your browser has it cached and colourising takes about thirty seconds of your own device's compute, whether you come back in an hour or next month.
Ordinary photographed scenes: portraits, family groups, landscapes, streets, food, animals. The model is weakest on close-ups of man-made objects with arbitrary colour, on heavily damaged scans, and on images that were never photographs — documents, line drawings and diagrams have no scene for it to reason about.
Yes — a colour strength control lets you push the predicted colour up or dial it back toward a tint. Because the strength is applied while the colour is recombined, changing it re-runs the model rather than updating instantly, and the page tells you that instead of pretending it is free.
The page detects that and warns you. Colourising a colour photo replaces the colours it already has with predicted ones, which is occasionally an interesting effect and usually not what you wanted.
Colourising does not repair damage. Use the AI Object Remover to erase scratches and tears, and AI Photo Restoration to rebuild faces and correct fading — one click hands the photo between them, without ever uploading it.

Stay Updated

Get the latest tools, AI features, and product updates. No spam.