Colorize Black & White Photo
Bring an old black-and-white or grayscale photo to life with AI-estimated colour. The model runs entirely in your browser — your photo is never uploaded.
Colorization runs in your browser — your photo is never uploaded.
How it works
This tool runs the Colorful Image Colorization model (Zhang, Isola & Efros) through ONNX Runtime Web on your own device. When you click Colorize, the model is fetched once from this site and cached, and the runtime is loaded from a CDN. Your photo is then processed locally — never uploaded. The network looks at the lightness of the image and predicts the two colour channels; we keep your photo’s original brightness at full resolution and lay the predicted colour over it, converting back to a normal RGB image you can download. Where your browser supports WebGPU, the maths runs on your graphics card; otherwise it falls back to WebAssembly on the CPU. Either way, the heavy model loads only when you run it.
How to colorize a black and white photo
- Drop a black-and-white or grayscale photo, or click to choose one.
- Click Colorize and wait while the model downloads on first use, then runs.
- Review the colorized result — the colours are an AI estimate.
- Download the colorized image as a PNG.
What to expect from the colours
Colorization is informed guesswork. The model has seen millions of colour photos and learned what usually goes where — blue for skies and water, green for foliage, believable skin tones — so the result generally looks natural and is a lovely way to reimagine an old picture. What it cannot do is recover the true colours, because that information was never captured in a black-and-white image: a specific jumper, car or flower might come out a different colour than it really was. Clear, well-exposed photos with recognisable subjects colorize best; very dark, very faded or heavily textured images give the model less to work with. Always treat the output as a pleasing estimate.
Why colorize in your browser
The photos people most want to colorize are personal — grandparents, old family gatherings, archive pictures — and uploading them to an unknown service is exactly what you would rather avoid. Running the model in your own browser tab removes that risk: the picture never leaves your device, there is no watermark and no queue. The honest trade-off is the first-run download: the colorization model is about 123 MB, fetched once and cached, and on devices without WebGPU the work falls back to the CPU, which is slower on large images. Throughout, your photo stays with you. Want to sharpen or enlarge the result afterwards? Try the upscaler, or convert an old scan first with RAW to JPG.
Related tools
Frequently asked questions
Is my photo uploaded to colorize it?
No. The colorization model runs in your browser with ONNX Runtime Web, using your device’s GPU through WebGPU where available and falling back to WebAssembly otherwise. Old family photos are exactly the kind of private images you would not want on a server, and here they never leave your device. Only the model is downloaded, once, then cached.
How accurate are the colours?
They are an educated guess, not a record of the real colours. The model was trained on millions of photos and predicts plausible colour from the shades of grey — skies tend to come out blue, grass green, skin in natural tones. It cannot know that a particular dress was red, so treat the result as a tasteful estimate rather than the truth.
Why is the first run a large download?
The colorization neural network is about 123 MB, and the ONNX Runtime it needs is a few more. They are downloaded once, the first time you colorize, and then cached so later runs start quickly and work offline. It is a bigger first download than our other tools because a colorization model is genuinely large.
Does the output keep my photo’s full resolution?
Yes. The model predicts colour at a small internal size for speed, but we keep your photo’s original brightness at full resolution and only stretch the predicted colour over it. The detail and sharpness come from your original image; the model adds the hue. Very large images take longer because the final recombination is done pixel by pixel.
Can I use the colorized image commercially?
Yes. The tool is free and the colorization model — “Colorful Image Colorization” by Zhang, Isola and Efros — is released under the permissive BSD-2-Clause licence, which allows commercial use. The image you download is yours to use.
Image Tools colorizes photos entirely within your browser. Your photo is never uploaded, and the colours produced are an AI estimate. The first run downloads the open-source colorization model (about 123 MB), which is then cached for offline use.