The short version: Palette.fm, MyHeritage's Photo Colorizer and Hotpot.ai are the three free tools worth your time in 2026, and ChatGPT or Google's Gemini image model will do a decent job too if you just ask nicely. None of them give you the "real" colours from the original scene, because that information was never captured on black and white film in the first place, so treat every result as an educated guess dressed up as history.
The free tools that work
There are dozens of "AI colourise" sites that show up when you search this, and most of them are the same wrapper around the same open-source model with a watermark bolted on. Here are the ones I'd point a family member towards.
Palette.fm
This is a browser-based tool that lets you type a text prompt alongside the upload, so you can say "1960s Kodachrome, warm tones" or "muted, overcast day" and it nudges the colour palette in that direction instead of leaving everything to chance. The free tier caps you at a handful of generations a day before it asks you to pay, and the output resolution is fine for screens and small prints but starts to soften if you try to blow it up to A4 or bigger.
MyHeritage Photo Colorizer
Built into the MyHeritage family history site, this one is aimed squarely at genealogy photos, which means it tends to do a better job on faces, skin tone and old fashioned clothing than general-purpose tools that were mostly trained on landscapes and street scenes. You need a free account, and it will try to upsell you into a subscription for "photo enhancement" bundles, but the core colourise function on a single image doesn't need a card.
Hotpot.ai
Fast, no-fuss, and the free version works well for casual, low-stakes photos where you just want to see grandad in colour rather than produce an archive-quality file. Expect a modest resolution cap and occasional odd choices, particularly with hats, uniforms and anything metallic.
ChatGPT and Gemini
If you've got access to GPT-4o or GPT-5 image generation inside ChatGPT, or Google's Gemini image model (the one people nicknamed Nano Banana), you can just upload the photo and ask it to colourise it realistically while keeping every face, pose and background detail identical. This works better than most people expect, and it's free within your normal daily usage limits. If you want a starting point for wording these requests well, I've put together a full set of copy-paste Gemini image edit prompts and a separate list of ChatGPT photo editing prompts that work, both of which include colourisation-specific phrasing you can lift directly.
A worked example: the 1962 wedding photo
Say you've got your grandmother's wedding photo from 1962, taken outside a church in a market town, slightly faded, with a visible crease down the middle from decades in an album. Here's the sequence that gets the best result without spending a penny.
- Scan or photograph it at the highest resolution your phone camera allows, in flat, even light with no flash, so there's no glare across the surface.
- Upload the raw scan to Palette.fm first and try the prompt "1962 wedding, natural skin tones, soft daylight" to see what it produces.
- Run the same original scan through MyHeritage's colourizer separately, because comparing two outputs side by side tells you more than trusting either one blindly.
- Zoom into the dress, the flowers and any medals or brooches, because this is where AI colourisers most often guess wrong, turning a cream dress slightly yellow or a red carnation into pink.
- Ask a relative who remembers the actual day, or was told about it, which version looks more plausible, rather than assuming the AI knows.
- Only then decide whether it's worth £5 to £15 on a paid tool like VanceAI or a Photoshop pass for a printable, high-resolution final version.
That last step matters more than people expect. Free tools are fine for sharing on WhatsApp or a family Facebook group. They start falling apart the moment someone wants to print it and frame it, because most free tiers cap output somewhere between 1000 and 2000 pixels wide, which looks soft and blocky at anything larger than a 6x4 print.
What "free" costs you
Every free colourisation tool is free for a reason, and it's usually one of these three: a watermark stamped across the image, a resolution ceiling low enough to nudge you toward the paid tier, or a daily limit that resets you back to zero after two or three photos. Hotpot and Palette.fm both do the resolution thing. Some of the smaller DeepAI-style demo tools do the watermark thing and don't tell you until after you've uploaded a precious family photo.
The workaround that costs nothing but time is running the same photo through two or three free tools and keeping the best output from each, rather than assuming any single tool is the "right" one. If you're already comfortable editing images, Adobe's Neural Filters inside Photoshop include a colourise option too, and I've covered how that sits alongside other AI features in this rundown of AI tools inside Photoshop, worth a read if you're going to be doing this more than once.
The bit nobody wants to say out loud
Colourisation is not restoration. It's a plausible reconstruction based on statistical patterns learned from millions of colour photos, and it has no idea what colour your great-grandad's coat was. It's guessing based on what coats of that era in that kind of photo usually looked like. Historians and archivists have argued about this for years, because colourised versions of famous historical photographs get shared as if they're documentary fact when they're closer to an artist's impression. The Imperial War Museum and similar institutions have been careful to label colourised images as interpretations rather than records, which is a distinction worth keeping in mind before you post a "restored" family photo captioned as if it's now historically accurate.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
This matters more with skin tone than with almost anything else. AI colourisers trained mostly on modern photo datasets can shift skin tone in ways that don't match the person at all, and this shows up disproportionately in older photos of people with darker skin, where the model has less to go on from a faded, low-contrast black and white image. If a colourised result looks wrong to someone who remembers the person, believe them over the software every time.
Which free tool suits which photo
- Faces and family portraits: MyHeritage Photo Colorizer, because it's tuned for exactly this.
- Landscapes, street scenes, group shots outdoors: Palette.fm, because the text prompt gives you some control over lighting and mood.
- Quick one-off, low stakes, just curious: Hotpot.ai or ChatGPT, both fast and free.
- Damaged, torn or heavily creased originals: none of the above will fix the damage first, so repair the tear or crease with a photo restoration tool or a prompt-based edit before colourising, not after. If you want prompt wording for that kind of repair-then-colourise workflow, the list of ChatGPT photo editing prompts trending on Instagram has several aimed at exactly this.
If you're doing a whole batch of old photos at once, say for a milestone birthday slideshow or a family history project, it's worth setting aside an afternoon rather than trying to squeeze it into ten minutes, because the free daily limits on most of these tools will slow you down if you rush it.
When it's worth paying
Paying makes sense in three situations: you need a print-resolution file above 3000 pixels wide, you're colourising more than about ten photos and the free daily caps are wasting your afternoon, or the free result keeps getting a specific detail wrong (a uniform colour, a flag, a car) and a paid tool with manual colour correction lets you fix it by hand. For a single photo for a Christmas card or a family group chat, free is enough. For a wall print or a wedding anniversary gift, spend the £6 to £15 and get it right.
And if you're planning to do this kind of AI photo work regularly, whether for family history or for content, it's worth thinking about the same prompts you're using for colourisation in other seasonal edits too. My Thanksgiving photo prompts for ChatGPT cover a similar family-photo-editing approach that transfers directly to old black and white shots once you swap the prompt wording.
Frequently asked questions
Is there a completely free way to colourise old photos with no watermark?
Yes. Palette.fm, MyHeritage's Photo Colorizer and ChatGPT or Gemini image generation all currently produce watermark-free results on their free tiers, though each caps you at a limited number of images per day and a modest output resolution.
Which free AI tool is best for old family photos specifically?
MyHeritage's Photo Colorizer tends to handle faces and skin tone more sensibly than general-purpose tools, because it was built specifically for genealogy photos rather than trained mostly on landscapes and general photography.
Are AI colourised photos historically accurate?
No. They're plausible guesses based on patterns learned from modern colour photos, not a record of the actual colours in the original scene, so they should be treated as an interpretation rather than a fact, especially around skin tone, fabric colour and uniforms.
Can I colourise a torn or badly damaged old photo?
Yes, but repair the tears and creases first using a restoration tool or a targeted AI editing prompt, then run the repaired scan through a colouriser, because colourising a damaged original usually just produces a colourful version of the same damage.
