Artificial Intelligence has revolutionised how we edit images, raising the big question: Can you colourise old photos with AI for free online?

The short answer is yes, but with significant caveats. While AI tools like MyHeritage and Photoshop’s Neural Filters can provide a quick, impressive “guess” at colours, they lack the historical knowledge and artistic nuance of a human restorer.

In this guide, we test the top AI colourization tools, expose their limitations, and explain when you should rely on an algorithm—and when you need a professional artist.

How AI Photo Colourisation Works

AI colourization uses “Deep Learning.” The software has been trained on millions of colour photographs to recognise objects (like trees, skies, and faces) and predict what color they should be in a black-and-white image.

For casual use, this is fantastic. However, the AI is essentially “guessing” based on averages, not specific historical facts.

Review: Top AI Colorizer Tools Tested

We put the most popular tools to the test using complex images involving foliage, brickwork, and people.

1. MyHeritage In Color™

• The Verdict: Currently one of the market leaders for quick fixes.
• Pros: Handles skin tones reasonably well; easy to use.
• Cons: Struggles with random purple “bleeding” colours and colour bleed (e.g., green from trees spilling onto people). It often misses finer details. Purple ingress bleed randomly. Favour blue black tones.
• Test Result: In our garden test photo, MyHeritage improved over the years but still failed to distinguish between the woman and the bushes, missing her entirely in some passes.

2. Photoshop Neural Filters (Colorize)

• The Verdict: Surprisingly inconsistent.
• Pros: Integrated directly into your workflow if you own the software.
• Cons: Our tests showed it effectively “gave up” on complex scenes, creating a muddy mess where it couldn’t distinguish between plants, buildings, and people.
• Advice: Avoid for complex scenes; acceptable for simple portraits.

3. Colourise.sg / DeepAI

• The Verdict: A strong contender for quick online results.
• Pros: Often identifies main subjects well.
• Cons: Struggles with grime or dirt on faces, interpreting it as skin pigment rather than surface dirt.

4. Flux and Nano Banana

• The Verdict: Modern GenAI tools like Flux and Nano Banana are fantastic at generating beautiful, hyper-realistic images from text prompts.

The 5 Major Flaws of Modern AI Colorization

Generative AI has evolved past the muddy, bleeding colors of 2018. Today’s state-of-the-art models (like Flux or Nano Banana) can flawlessly generate a “man with brown hair, a blue striped coat, and a gold pocket chain” simply by typing it into a prompt box.

But when it comes to preserving a genuine piece of family history, the sheer power of these new models is exactly where the problems begin. Here is why an algorithm still cannot replace a human professional:

1. The “Prompt Reliance” Problem: AI Invents, You Guess Modern AI will execute your instructions perfectly—but you have to know what to tell it. If you don’t know the exact color of a 1920s storefront sign or a specific Victorian dress material, the AI will just make a highly realistic, beautiful guess based on what looks visually pleasing. A professional doesn’t guess; they investigate.

2. Historical Inaccuracy This remains the biggest danger for genealogists and historians. An AI might expertly render a soldier’s uniform a crisp, realistic blue because you prompted it to, or because it defaults to standard military colors. But it cannot check regional regimental archives to see if that specific 1915 uniform should actually be khaki green. It doesn’t research the specific dyes available in the 1890s vs. the 1950s. It only knows what is in the prompt, not what is in the history books.

3. The “Clean Up” Factor: Embedding the Damage AI struggles to understand context. It cannot tell the difference between a natural mole on a cheek, coal dust on a miner’s face, or a physical speck of dirt on the photograph itself. AI will often colorize the dust or grime as if it were skin pigment, permanently embedding the damage into the final image. A professional human eye cleans and restores the physical damage before applying color.

4. Hallucination and the Loss of Likeness Older AI simply overlaid color; modern Generative AI fundamentally redraws the image. When faced with a blurry face or a faded eye, the AI will “hallucinate” (invent) new pixels to make the image look sharp. While the result looks stunning, it might change your great-grandfather’s jawline, alter the shape of his eyes, or add dimples that were never there. You are no longer looking at your ancestor; you are looking at an AI-generated lookalike.

5. Destructive “Inpainting” vs. Archival Repair When a photo is severely torn or scratched, modern AI uses a technique called “inpainting” to guess what was underneath the damage. Rather than meticulously repairing the tear, the AI often deletes the damaged area entirely and overwrites it with a newly generated background. A professional performs non-destructive, archival-grade repairs, carefully stitching the original details back together to preserve the true artifact.

Comparison: Generative AI vs. Professional Human Restoration

We tested an image of a boy in a garden to compare the results directly.

The Case Study Result: When applying AI to a line drawing (a teddy bear), the AI failed to understand the context of the art. When applied to a photo of a garden, the AI painted the ground “brown” (correct) and the leaves “green” (correct), but it couldn’t handle the complexity. It colored the brick walls with green patches and failed to identify the specific types of plants.

The Human Touch: A human artist identifies that the “dirt” on the boy’s face is grime, not skin color. We identify that the vine is a specific species that should be a different shade of green than the grass. We ensure the brickwork matches the local architecture of the era.

FAQ: Common Questions About AI Colourisation

Is there a free AI that colourises photos perfectly?

No. Tools like Palette.fm, MyHeritage, and Hotpot.ai offer free tiers, but none produce “perfect” results. They are best used for visualisation, not preservation.

Can AI fix damaged photos while colourising?

Generally, no. Most AI colorizers layer colour over the damage (scratches, tears). You should always digitally repair and restore the photo (remove scratches/tears) before attempting to colourise it.

Why does my AI-colored photo look purple or muddy?

This is a common artefact called “chroma noise” or low-resolution output. AI often works at a lower resolution to save processing power, resulting in muddy, undefined colours.

Conclusion: When to Hire a Pro

AI is a fun tool for a quick “what if” look at your ancestors. It brings a moment of life to a black-and-white memory.

However, if you are looking to archive a family treasure, print a large canvas, or ensure historical accuracy for a book or project, AI is not enough. You need a human eye to research the era, understand the lighting, and meticulously paint every pixel.

Ready to see your photos in true, accurate color? Check out our Colorizing Services or view our Portfolio to see the difference professional restoration makes.