How to colorize black and white photos

Updated 13 October 2026 · UnBlur AI guides · 6 sources

AI results are estimates. AI restoration predicts what a sharper or repaired photo might look like. It can invent or change details, especially faces, text and fine patterns. Always keep your original, and do not use an AI-restored image as evidence or to identify someone.

You can colorize a black-and-white photo by painting color onto it in a photo editor or by letting an AI tool predict the colors. Both approaches can look convincing. Neither can know the real colors, because a black-and-white photo only records how light or dark each area was. A red dress and a green dress of the same brightness look identical in gray.

That is why the computer vision research on automatic colorization describes the task as producing a plausible color version of a photo, and calls the problem underconstrained ¹. Treat any colorized photo, including one you made yourself, as an interpretation.

Before you start

1. Scan the original well. Colorizing a poor copy multiplies its flaws. See how to scan old photos properly.
2. Fix tone first. Set black and white points and contrast so skin, fabric and sky have good separation ⁴. Faded prints are covered in restoring faded photos.
3. Repair damage first. Scratches and spots pick up odd colors. See repairing damaged photos.
4. Keep the black-and-white master. Archival practice keeps an unaltered master image and never applies color or tone restoration to it ².

Research the real colors

The more you know, the less you have to guess. Useful clues include:

• Family memory. Relatives may remember a car, a house color, a favorite coat or someone's hair and eye color.
• Writing on the back or in letters and diaries from the time.
• Uniforms and insignia. Military, school, sports and work uniforms often followed documented color rules. Museums and archives hold examples.
• Other color photos from the same family, place or decade. Color slides and prints of the same house or town are the strongest evidence.
• Things that do not change. Landscapes, buildings and some plants can be checked against the place today, keeping seasons in mind.

Write down where each color came from. If you share the photo later, those notes show what is known and what is guessed.

Colorizing by hand

Most layer-based photo editors, including free desktop ones, can colorize a photo by hand:

1. Convert the scan to RGB color mode so it can hold color.
2. Add a new layer for each area, such as skin, hair, clothing and sky, and set its blend mode to Color so the brightness of the original shows through.
3. Paint with low-saturation colors. Real skin and fabric in old scenes are rarely bright.
4. Vary skin tones slightly: cheeks, lips and ears are a little warmer than the forehead.
5. Leave some areas nearly neutral. Overcolored photos look like postcards.

Hand colorizing is slow, but every choice is yours and can be based on your research.

Colorizing with AI

AI colorizers learn typical colors from large collections of color photos (one widely cited research model was trained on over a million) ¹: skies are usually blue, grass usually green, skin within a range of tones. Research on automatic colorization aimed at results vivid enough that people might take them for real color photos ¹. That is exactly the risk: a colorized photo can look authentic while showing the wrong uniform color, the wrong eye color or an invented pattern.

In UnBlur, the Colorize mode ("Add color to B&W") takes a black-and-white photo from your library and returns a colorized version of about one megapixel after roughly 30 to 60 seconds. The model is instructed to keep each person's face, expression, pose and the composition unchanged, but it is a generative model ⁵, so check details and treat the colors as suggestions. If you know a color was different, an editor's hue tools can correct it afterward.

Check and label the result

• Compare faces with the black-and-white original at the same size: generative tools can alter features ⁶.
• Look for color bleeding across edges, such as skin tone spilling onto a collar.
• Label shared copies, for example "Colorized in 2026; colors are estimated." Put the note in the caption or the file's description, along with who, what, where and when ³.
• Share the black-and-white original alongside it, so no one mistakes the colorized version for a color photograph.

For more on where AI goes wrong, see what AI photo restoration can and can't do.

Common questions

Sources and further reading

1. Zhang R, Isola P, Efros AA. Colorful image colorization. ECCV 2016 (Lecture Notes in Computer Science). arxiv.org
2. Federal Agencies Digital Guidelines Initiative. Technical Guidelines for Digitizing Cultural Heritage Materials, 3rd edition (2023), section 3.8 Prints and Photographs (PDF). digitizationguidelines.gov
3. U.S. National Archives. Digitizing Family Papers and Photographs. archives.gov
4. Apple Support. Apply levels adjustments to a photo or video on Mac. support.apple.com
5. Black Forest Labs, et al. FLUX.1 Kontext: Flow matching for in-context image generation and editing in latent space. arXiv:2506.15742. 2025. arxiv.org
6. Blau Y, Michaeli T. The perception-distortion tradeoff. IEEE CVPR. 2018. arxiv.org