Your detail is untouched
Only the colour channels are predicted; the original luminance is preserved exactly, so grain, sharpness and tonal range come through unchanged.
Free Tool
AI Photo Colorizer
Turn black & white and sepia photos into colour. The realistic mode runs the DDColor neural network entirely in your browser, it studies the scene and picks plausible, natural colours for skin, sky, foliage, and fabric. Prefer a look instead of realism? Four instant palette styles (vintage, sepia, saturated, pastel) apply with no download at all. Compare every result against the original with the built-in slider. Your photos never leave your device.
DDColor neural colourInstant palette stylesBefore/after sliderNo media uploadFree to use
Result proof
Colour you can compare against the original luminance
01
Input
Black-and-white source
02
Local pass
DDColor or named palette
03
Result
Separate colour PNG
How to verify it: Drag the comparison slider across skin, signage, uniforms and painted objects; those arbitrary colours need source-based correction rather than trust in a confident prediction.
Plausible, not documentary
Colour is inferred, it is not recovered
A black-and-white photograph does not contain its colours; they were never recorded. A colouriser predicts what is most likely given everything it has seen, which produces convincing skies, skin and foliage and can be confidently wrong about a dress or a painted door. Knowing which is which is the whole skill of using it well.
Luminance kept
Detail preserved
Strong on the common cases
Skies, vegetation, wood and skin are consistently well predicted because they vary within narrow, learnable ranges.
Weak on the arbitrary
Clothing, paint, vehicles and signage are genuine coin flips, the model has no way to know, and it will not tell you it is unsure.
Know before you process: Treat the result as an interpretation. For archival or historical work, verify colours against a written record rather than trusting a plausible-looking output.
What people use this for
Family photographs
The most common reason by far, and the one where the result matters most emotionally. Skin, hair, foliage and sky are the model’s strongest categories, so a portrait taken outdoors usually comes back convincing. Clothing is the coin flip, if you know what colour a dress was, the palette controls are there for a reason.
Local and community history
Streetscapes and building exteriors colourise well because brick, stone, timber and vegetation are consistent. Painted signage and vehicles are not, treat those as illustration rather than record, and say so if the image is going on display.
Restoring before printing
Because only the colour channels are predicted and the original luminance is preserved exactly, a colourised scan keeps the grain and tonal range of the original. It will print with the same detail the black-and-white version had.
Design and moodboards
Turning archive imagery into something that sits alongside modern colour photography. Here plausibility is the whole requirement and accuracy does not apply, which makes it the lowest-risk use of the tool.
What to be careful about
- The colours are guesses, and confident ones. The photograph never recorded them. The model predicts what is statistically likely, and produces an equally polished-looking image whether it is right or completely wrong.
- Arbitrary colours are unknowable. Clothing, paint, cars, flags and signage could have been anything. If a specific colour matters (a uniform, a livery, a flag) check a written source and correct it by hand.
- Faded or damaged scans mislead it. Sepia toning and yellowed paper are read as image content. Convert to neutral grayscale first and the result improves noticeably.
- Do not present the output as a historical record. For archival, journalistic or memorial use, label it as a colourised interpretation. That is a courtesy to whoever sees it next, who has no way to tell.
How it works
- Drop a black & white, sepia, or faded photo (JPEG, PNG, or WebP)
- Realistic mode runs the DDColor model in your browser, a one-time ~460 MB download you approve first, cached for offline reuse
- Style modes (vintage, sepia, saturated, pastel) colour-grade instantly with zero download
- Drag the slider to compare, then download the colorized PNG
Frequently asked questions
How accurate are the AI colours?
DDColor predicts plausible colours from what it learned across millions of photos, grass becomes green, skies blue, skin tones natural. It cannot know the true historical colour of a specific dress or car, so treat the result as a faithful-looking interpretation, not a record.
Why is there a 460 MB download for realistic mode?
The DDColor network weights run entirely on your device, that is what keeps your photos private. The download happens once, is cached by your browser, and works offline afterwards. If you skip it, an instant classical pass colorizes with a lighter touch instead.
Does it work on damaged or very old photos?
Yes, colorization works on any grayscale input. For heavily damaged photos, results improve if you first run the image through the upscaler or clean it up in the editor, then colorize the restored version.
Can I colorize a photo that already has some colour?
Yes. Faded or colour-cast photos are re-interpreted by the model; stylised modes apply their palette over whatever colour exists. For fully coloured photos the realistic mode makes only subtle changes.
Does the image get uploaded?
No. Both the realistic model and instant palette modes process the selected pixels in this browser tab.
Related tools
Ready to try it?
Free, private, no signup.
Colorize a photo →