Colorize Black and White Photos

A black and white photograph asks you to imagine the room it was taken in. Colour gives it back — the shade of a coat, the green of a garden, the warmth of skin — and with it a sense that the moment really happened rather than belonging to an archive. For family history that difference is the whole point: relatives recognise faces in a colourised portrait that they had walked past in monochrome for years.

What colour brings back

• Recognition: faces read as people rather than as records, which is what makes a family photograph worth sharing again.
• Depth: colour separates a subject from a background that greyscale had flattened into one tone.
• Dating: the colour of clothing and paintwork often places a photograph within a decade when nothing else in it does.
• Repair: a colour print from the 1980s that has drifted pink can be pulled back toward the colours it was printed with.

Colour is added by a neural network that runs on your own device. The photo is never uploaded.

1. 1Add your photo: Drop in a black and white scan or photograph — JPG, PNG, WebP, AVIF, BMP or GIF.
2. 2Press Colorize: The first run downloads the colour model once, then works offline from then on.
3. 3Adjust the result: Move colour strength and saturation until it looks right. Both update as you drag.
4. 4Correct anything wrong: With Pro, pick a colour and click the photo. The colour spreads across whatever you clicked.
5. 5Download: Export at the photo's original resolution as PNG, JPG or WebP. No watermark, on any plan.

The tool does not tint your photo. It runs a neural network that was trained on more than a million photographs, learning which things in a picture tend to be which colour — that skin sits in a narrow band of warm tones, that a large bright region above a horizon is usually sky, that dense texture at ground level is usually foliage. Given the brightness of your photograph, it predicts a colour for every part of it, and different parts get different answers. That is the difference between colourising and tinting: a tint applies one colour scheme to everything, and cannot tell a face from a field.

The prediction is combined with your photograph's own brightness at its original resolution, so every grain, scratch and detail in the scan survives untouched — only colour is added. This is also why output is always full size and why there is no resolution charge here: the recognition step costs the same on a 12-megapixel scan as on a thumbnail.

Colour is predicted once per photograph and then combined with your image at its original resolution using your device's graphics hardware, so the strength and saturation controls redraw as you drag them even on a 12-megapixel scan.

These are two different jobs and the tool tells them apart automatically. A true black and white scan has no colour to keep, so the prediction is used as it stands. A faded colour print is the opposite case: it still holds the real colour of everything in the frame, just weakly, and that surviving colour is the only ground truth in the picture. So where your photograph still has colour, its own hues are trusted and simply restored to strength; where it has none, the prediction fills in. If the automatic choice is wrong, you can set it by hand.

• Detect: measures the colour left in the photograph and picks for you. Correct on almost everything.
• Black & white: force a full colourisation, for a scan whose paper has yellowed enough to look like colour.
• Faded colour: keep the photograph's own hues and bring them back to strength, for prints that have drifted.

No colourizer knows that a particular dress was red. It has no way to — the information is simply not in a black and white photograph, so any tool that gets it right has guessed well rather than known. What matters is whether you can correct it when it guesses wrong, and most tools give you nothing but a slider.

With Pro you pick a colour and click the thing you want that colour to be. The point is dropped into the network itself rather than painted over the top, so it spreads along the object you clicked and stops where that object stops — a coat becomes red, and the wall behind it does not. Colour strength and saturation stay available on every plan, and both redraw as you drag them because the recognition step is only done once.

The colour model is downloaded to your device the first time you use the tool, and everything after that happens where the photo already is. Nothing is uploaded, nothing is queued on a server, and no copy of your family photographs exists anywhere but on your own machine. That is not a policy promise — there is no upload path in the tool to make a promise about.

It is also why there is no watermark and no credit system. Hosted colourizers meter you because every colourisation costs them a server; this one costs us nothing after the model has downloaded, so free means full resolution, unwatermarked, as many times as you like.

Four jobs account for most of it.

• Family history Portraits and wedding photographs from the 1900s to the 1960s, colourised so living relatives recognise the people in them.
• Restoring faded prints Colour prints from the 1970s and 1980s whose dyes have drifted, pulled back toward what they looked like when new.
• Local and school archives Scanned collections where colour makes buildings, uniforms and streets legible to people who never saw them.
• Illustration and reference Historic photographs used as colour reference for painting, model making or set design.

• Image enhancer
• Image upscaler
• Background remover
• Image converter