Colorize old family photos without turning them into paintings

What AI does when it colorizes a vintage photograph

A black and white photo stores one thing per pixel: brightness. The color is gone, and nothing in the grey values says whether a dress was red or green, because both can come out as the same shade of grey. When you colorize a vintage photograph with AI, the model does not recover that lost information. It recognises what is in the picture and paints each region with the color it has seen most often for that kind of thing.

In practice that means the model is reliable where the world is predictable and guessing where it is not:

• Usually right: skin, sky, grass and leaves, tree bark, wooden furniture and floors, brick, stone, sand, water. These have a narrow range of likely colors, and the model has seen millions of examples.
• Plausible but invented: clothing, curtains, wallpaper, cars, painted doors, flowers, toys. A cardigan becomes beige, navy or burgundy because those are common, not because your grandmother wore one of them.
• Close to a coin toss: eye color, the exact shade of hair, uniforms with specific regimental colors, signs and packaging.

This matters for family photos more than for any other kind. Relatives will look at the result and say "her coat was green, not brown" or "grandad had blue eyes". The AI cannot know that, and no setting will make it know. If one detail really matters (a wedding dress, a uniform, the family car), treat the colorized version as an interpretation and say so when you share it. Write the real colors on the back of the print or in the file name, so the next generation does not mistake the guess for a record.

The model also works from the brightness of the scan, not from its tint. A sepia or yellowed print is read as greyscale, so the brown cast of the old paper does not carry into the colors. That is usually what you want: the result looks like a color photograph rather than a tinted one.

Clean the scan before you add color to a black and white photo

The most common reason a colorized family photo looks fake is not the colorizer. It is the dust, grain and cracks that were on the scan when it went in. The model treats every speck as part of the picture: a scratch across a cheek is read as an edge, a cluster of film grain as texture, a stain as an object. Color then clings to those defects. You get a pinkish blotch on a sleeve, grey-green patches on a forehead, or a sky that is blue in some places and lilac in others.

Removing those defects first gives the colorizer smooth, honest areas to work with, and the color comes out even. The order matters:

1. Grain and noise first. Old prints made from small negatives, photos taken indoors, and scans of newspaper or magazine pages carry heavy grain. Use a tool to remove noise from a photo with a moderate intensity: enough to smooth the wall behind the people, not so much that faces turn waxy.
2. Damage and faces second. Then restore old photos: this step rebuilds faces that have gone soft, cleans and enlarges the background, and can repair scratches and creases you mark with a brush.
3. Color last. Only now does the picture go into the colorizer.

Doing it the other way round (color first, then clean) does not work. Once a blotch has been colored, the cleaning tools see a colored patch, not a defect, and keep it.

A note on the source. If all you have is a copy of the photo that someone sent over a messaging app, the compression has already destroyed fine detail and added its own block patterns. The cleaning steps help, but they cannot bring back what was thrown away; our article on the pixelated WhatsApp photo explains what can and cannot be recovered. Whenever possible, borrow the original print and scan it yourself. Our guide to scanning slides and negatives at home covers film, and a flatbed scanner at a high resolution is enough for prints.

Natural or vivid, and how much intensity

The colorizer in Airpic gives you two modes and one slider. They answer different questions, so it is worth understanding both before you press the button.

Natural keeps colors restrained: skin in muted, believable tones, clothing in the slightly desaturated palette of real photographic prints from the 1940s to the 1960s. Vivid pushes saturation and contrast in the color, closer to modern digital photos. As a rule:

• Use natural for portraits, family groups, weddings, christenings and anything where faces are the main subject. Skin is where viewers notice exaggeration first, and vivid tends to push it towards orange or pink.
• Use vivid for landscapes, streets, beaches, gardens and postcards, where the subject is the place rather than the people and a stronger sky and greener trees look right.

The intensity slider controls how much of the predicted color is mixed into the original. At the top of the range you get the full prediction; lower down, more of the original grey tones show through, and the result looks like a carefully hand-tinted print.

For family portraits the credible result is usually below the maximum. The reason is simple: the full prediction is confident everywhere, including in the places where it is guessing (clothes, eyes, background objects). Reducing the intensity makes the guessed areas quieter while skin and sky stay recognisable. A practical method: start in the middle of the slider, raise it until the faces look alive rather than grey, then step back a little. If the cheeks look like make-up or the lips are too red, you have gone past the point.

Two more signs that you need less color, not more: a white shirt or wedding dress that picks up a tint, and hair that turns uniformly brown or orange across several people. Both mean the model is filling uncertain areas with its favourite answer.

When black and white is the better choice

Colorizing does not always improve a photo, and part of doing it well is knowing when to leave the original alone. These are the cases where the grey version usually wins:

• Studio portraits with worked light. Many portraits from professional studios were lit for black and white: a strong key light, deep shadows, a glowing rim on the hair. That modelling was the photographer's whole design. Color competes with it, and the result often looks flatter than the original.
• Very high-contrast portraits. When half the face sits in deep shadow, there is little brightness information left there. The colorizer has nothing to work with, so the shadow side tends to go muddy brown or grey-green.
• Documents and photos with text. Certificates, letters, passports, ration books, newspaper clippings and school reports gain nothing from color. Worse, a colorized document can look like a fake. Keep them grey and, if they are hard to read, improve them in other ways instead.
• Photos used as evidence. If you are building a family tree or dating a picture from the clothes and cars, the invented colors can mislead anyone who uses the image later.
• Prints that are badly damaged. Where a large part of the image is missing or stained, color amplifies the damage. Restore it first and decide afterwards; often the restored black and white version is the one people prefer.

There is no harm in trying: the colorizer works on a copy and the original scan stays as it was. Keep both versions in the family archive, with the black and white one as the reference.

If you want to go further with a photo you have restored, our article on how to animate old family photos with AI explains what can go wrong with faces and how to avoid it. And if your family photo is going to be printed large for a wall or an anniversary, read how to upscale an image for printing before you send it to the print shop.

How to colorize your grandparents' photo in Airpic, step by step

Here is the exact sequence we use for a typical family print: a portrait of grandparents from the 1950s, scanned on a flatbed, with some grain, a crease across one corner and soft faces. Each tool processes the image and shows you a preview for free; you only download the full file when you are happy with it.

1. Scan and check the file. Scan the print at a resolution that gives you a few thousand pixels on the long side. The tools accept JPG, PNG (including 16 bit), WebP, TIFF, HEIC and AVIF, apply the EXIF orientation and keep the ICC profile, so a phone photo of the print works too if you shoot it flat and without glare. Crop away the scanner bed before you start.
2. Remove the grain. Open the tool to remove noise from a photo, upload the scan and use a moderate intensity. Judge the result on a flat area such as the wall or the sky, then on the faces: you want the grain gone from the wall without the skin turning to plastic. If the scan is already clean, skip this step.
3. Restore the damage and the faces. Upload the cleaned image to the tool to restore old photos. Keep the scale at ×1 if your scan is already large (the tool accepts up to 20 megapixels) and use ×2 for small prints. Turn on face reconstruction and leave the restorer's own color option off: you will add color in the next step with more control. Use the fidelity slider to balance sharpness against resemblance; on very degraded faces a lower value keeps the person looking like themselves. If you have other photos of the same person, add up to 3 as references to guide the likeness. For the crease, tick the option to repair damage and paint over it with the brush. When the job finishes, use the stage chips to compare each step and check that nothing was invented.
4. Add the color. Open the photo colorizer, upload the restored image, choose natural for a portrait and set the intensity a little below the maximum. Drag the comparison slider across the faces and use the 100% view to check the edges: collars, hairlines and hands are where color spills first. If the skin looks orange, lower the intensity; if the picture still feels grey, raise it in small steps.
5. Save the right format. Download PNG for the family archive and JPG for sharing. Keep the original scan, the restored black and white version and the colorized one together, with the date and the real colors you know written in the file name.

Speed is not the bottleneck. In our tests on 19 September 2026, a 1,400 × 1,050 px black and white photo was colorized in 3 seconds. The time goes into choosing the mode and the intensity, and that is time well spent.

About the files: they are processed on GPUs in the European Union, deleted 1 hour after download and 1 day at the latest, and never used for training. There is no account to create. One full download a month is free; after that you use credits that do not expire, and the price is on each tool's page. Because each step above ends with a download, look carefully at the preview before you download at every stage.

For more detail on the color step itself, see our guide to colorizing black and white photos. When your scan is clean and restored, open the photo colorizer and give your grandparents' picture its color.

FAQ

Keep reading