Colorize Black and White Photos

Yes. No account, no credits, no watermark, and no daily counter. The site is funded by ads. The only limit is an automatic per-IP cap that blocks bots, and ordinary use never reaches it.

About 20 seconds is typical. Across our eight test photos the full round trip ran from 2.3 seconds to 62.8 seconds. The spread is cold starts: the GPU container scales to zero when nobody is using it, so the first photo after a quiet period pays for the boot and the ones after it do not.

No, and no tool can give you those. A black and white negative never recorded colour, so anything you get back is a prediction. The model is reliable on things it has seen millions of examples of, such as skin, foliage, sky, wood, and denim. It is guessing on a specific dress, a car's paintwork, or a uniform. Treat the result as a good-faith interpretation, not a restoration of fact.

No, and that is the point of the second half of our pipeline. We keep your photo's brightness channel exactly as you uploaded it and lay only the predicted colour over it. Measured across the test set, the mean brightness difference between input and output was 0.08 L*, which is the rounding error of writing the file.

Because we damp it on purpose. The model runs warm: on a test photo with a known-white dinner plate its colour error against the real original was 29.3 at full strength, 24.7 at 85%, and 20.7 at 70%. We ship 80%, which is where the orange wash is gone but the photo still clearly reads as colour rather than a tinted grey.

It is sent over HTTPS to our servers and to the AI provider that runs the model, because colorization cannot run in a browser. It is deleted automatically within 7 days, is never sold, never shared for advertising, and never used to train models. No account is required.

JPEG, PNG, WebP, or GIF, up to 7.5MB. Before upload your photo is resized so its longest side is 1024 pixels, which is also the size the coloured PNG comes back at. That cap is deliberate: the same model needs 0.11 seconds of GPU time at 1024 pixels and 33.9 seconds at 3072, and the difference is not worth the wait on a free tool.

Completely different jobs. The Colorize Image tool is a tint: it pushes every pixel toward one colour you pick, which is useful for grading and brand-coloured assets, and it runs in your browser. This page runs an AI model that looks at the scene and predicts a different colour for each region, which is what you want for an old black and white photograph.

DDColor (piddnad/ddcolor on Replicate), chosen in a five-engine bake-off on 26 July 2026, plus our own chroma-damping step that rebuilds the result on your original luminance at 80% colour strength. The comparison table above has the numbers behind that choice.

You can, but colorizing will not repair the damage, and colour tends to follow whatever the scratch did to the brightness. Brush out the marks with the free object remover first, colorize second, and enlarge with the upscaler last if you want a print.

It is built for greyscale input. On a sepia-toned print the model mostly neutralises the existing tone and predicts a new one, so you get a re-interpretation rather than an enhancement. Use it on genuine black and white photos.

No. There is no account, no email, no extension, and no app. Open the page, drop in a photo, and download the result.