How Photo Colorizer Works
Colour for black-and-white photos two ways: instant classical styles with zero download, or the DDColor neural network for photorealistic colour — your choice, consent-gated.
Try it free →Step-by-step process
- 1
Choose a style
Realistic uses DDColor, a neural network trained to predict plausible colour from luminance — skin, foliage, sky, and materials come out looking natural. Vintage, Sepia, Saturated, and Pastel are instant classical LAB-space treatments that need no model download at all.
- 2
On-device colorization
DDColor is a ~460 MB one-time download, so it never fetches without your explicit OK — decline and the classical path colours instantly instead. Inference runs entirely in your browser; the photo itself never leaves your device.
- 3
Compare and export
A before/after slider shows the original against the colourised result so you can judge the change honestly. Export as PNG at the source resolution. In the image editor, the same colorizer sits next to Grayscale and is fully undoable.
Frequently asked questions
Are the colours “real”?
No colorizer recovers the true historical colours — that information is gone. DDColor predicts statistically plausible colour from context, which is why grass reads green and skin reads natural. Treat it as a well-informed guess, not a restoration of fact.
Why is the AI model such a large download?
DDColor carries full image-understanding weights (~460 MB). It downloads once, caches on your device, and works offline afterwards. If you skip it, the four classical styles run instantly with zero download.
Can I colorize inside the image editor?
Yes — AI Colorize sits beside Grayscale in the editor's colour tools, auto-suggests when a black-and-white photo is detected, and supports full undo.
What input works best?
Clean scans with good contrast. Heavily faded or damaged photos benefit from a contrast fix first; very dark scans give the model little luminance information to work from.