PhotoTools.ai

Photo Colorizer

Colorize a black and white photo

Add natural, period plausible color to a monochrome photo, with skin tones kept believable.

Free to try without an account. Signed in, a run costs 4 credits.

A black and white portrait, colorized
A black and white portrait, colorized. Drag, or use the arrow keys, to compare before and after.

Try it

No photo chosen yet.

Drop a photo here

or

JPG, PNG, or WebP, up to 25 MB and 20 megapixels.

How it works

  1. Add a photo: drag it onto the uploader or pick one from your device.
  2. The model picks colors that fit the era and the scene, then paints them in without moving anything.
  3. Look at the result, then download it or run another photo through.

These models reconstruct rather than recover: they produce a plausible version of what was probably there, so results vary with the source photo. Run it again or try a different image if the first pass misses.

What to expect

The model reads the scene and paints plausible color into a monochrome photo, holding composition and faces exactly where they were. Skin, sky, foliage, wood, and common materials come out most reliably, because each has a narrow believable range. Everything with a free choice attached is a guess: a dress, a car, painted walls, a uniform, a flag. It has no way of knowing that your grandmother's coat was green, and it will commit to something confidently either way. Treat the output as a colored version rather than a record, and correct the one region you care about in an editor afterwards. A print that is very low in contrast, already toned sepia, or heavily damaged gives it a weaker signal to read.

Best results: A clean black and white scan with a decent spread of tones from black to white and no heavy damage.

Common questions

Are the colors historically accurate?
They are plausible, not researched. The model picks colors that suit the era and the scene, and anything it cannot infer, such as clothing or a painted wall, is a confident guess.
Can I tell it what color something should be?
Not from this page: a run is one pass with no per object instructions. Recoloring the one region that matters in any editor afterwards is faster than fighting it.
Will I get the same colors if I run it twice?
Not necessarily. These are generative models, so a jacket that came back brown once may come back blue on a second pass. That is worth knowing as a feature: rerun if the first choice bothers you.
Should I restore the photo before coloring it?
Yes, if the print is damaged. Run Old Photo Restoration first, because scratches and creases left in place are read as features of the scene and get colored along with everything else.
Does it work on sepia photos?
It does, though a sepia or cyanotype print carries a color cast the model has to work past, so results are less predictable than from a neutral black and white scan.

See all tools