How to colourise a black-and-white photo
Drop the photo on the page and press Colourise photo. Drag the divider to compare before and after, set the colour strength, warmth and saturation to taste, and save it at full size.
Open a photo
A scan, or a phone photo of a print. JPG, PNG, WebP, GIF, BMP and AVIF open, up to 24 megapixels. Sepia and faded colour photos work too: only their lightness is used.
Colourise
The first time, the AI model (DDColor large for Best colours, 231 MB, or tiny for Fast, 56 MB) and its runtime download once and stay on your device. The colours take a few seconds on a graphics card, or 5 to 40 seconds on a processor.
Tune it
Colour strength goes from grey (0 %) to stronger than the model's guess (150 %). Adjust adds warmth, saturation, contrast, auto levels for faded prints, and grain.
Save
As JPG, PNG or WebP, the same size as your photo.
How true are the colours?
They are a guess. A black-and-white photo keeps how light each thing was, not what colour it was: a red and a green of the same lightness look the same in grey. The model has learned from over a million colour photos which colours things usually have, so skies, grass, water and skin tones tend to be close. A dress, a car, a painted door or a uniform can come out in a colour it never had.
- For family photos, ask someone who was there, then correct with Warmth and Saturation
- For history, say the photo was colourised: the colours are an impression, not a record
Why no detail is lost
The model looks at a 512 × 512 copy of your photo and answers with colour only. That colour is then laid over your photo at its own size in the CIE Lab colour space, which keeps lightness and colour apart: each pixel keeps the exact lightness it had and only gains colour. Fine detail, grain and sharpness come from your photo, never from the model, and a 20-megapixel scan stays 20 megapixels.
| Setting | What it does |
|---|---|
| Colour strength | How much of the guessed colour goes on. Lower for a gentle, hand-tinted look; above 100 % for livelier colour |
| Restore faces | Before colouring, rebuilds faces in damaged or blurry photos with GFPGAN, then blends them back. Off by default: it can change how a face looks |
| Adjust | Warmth, tint, saturation, vibrance, exposure, contrast, shadows, highlights, sharpen and grain, on the full photo |
Where does the photo go?
It stays on your device. The AI model downloads from this site once and runs right in your browser, on your graphics card if it can, otherwise on your processor.
- Downloaded once, then kept in your browser for next time
- Saved without metadata like location or camera
Specs and browser support
| Input | JPG, PNG, WebP, GIF (first frame), BMP, AVIF, up to 24 megapixels. Colour and sepia photos are made grey first |
|---|---|
| Output | JPG (92 % quality), PNG or WebP at the photo's full size; transparency kept in PNG and WebP |
| Model | DDColor (ICCV 2023, Apache-2.0), weights stored as int8: Best colours is DDColor large, 231 MB; Fast is DDColor tiny, 56 MB. Plus the onnxruntime-web runtime (MIT), 14 or 17 MB. Face restoration: GFPGAN 1.4 and YuNet, 86 MB more |
| Speed | A few seconds on a graphics card with WebGPU; on a processor 5 to 40 seconds with Best colours, 5 to 20 with Fast. Faces add about 2 seconds each |
| Browsers | WebGPU in Chrome and Edge. Other browsers use the processor. A computer with 4 GB of memory or more is best |
| Price | Free, no watermark, no account |
Questions
No. The AI guesses likely colours from what it learned on millions of colour photos: skies blue, grass green, skin tones. It cannot know the colour of a dress or a car that day, so treat the result as an impression, not a record.
No. The AI model is downloaded to your browser and runs there. Your photo stays on your device.
No. The model only decides the colour. Every pixel keeps the lightness of your photo, at its full size up to 24 megapixels, so detail and grain stay as they were.
The first time, your browser downloads the AI model and its runtime: about 245 MB for Best colours, or 71 MB for Fast, plus 86 MB with face restoration. They are kept on your device, so the next photo starts at once.