The ability to convert sepia or black-and-white photographs into colour is not a new one, but thanks to artificial intelligence (AI) it’s a technique that can now be applied to any digital photo with just a few clicks. We tested three photo colourisation apps (MyHeritage, Colourise.com and Ancestry) using a set of eight images - a mixture of photos from across the 20th century. All of the tools we feature in this test - along with similar ones from Colorize (colorize.cc), DeepAI and Photomyne - are designed to be accessed directly from your web browser. In most cases, you simply upload your photograph, click a button and wait for the results to be delivered to you.
Some services provide additional tools. For example, after colourising a picture with MyHeritage, you can navigate to the ‘Photos’ section and select the image. Click the ‘Settings’ button above your image and you’ll be able to tweak the effect several ways via four options, one of which allows you to change the version of the underlying DeOldify model that was used. MyHeritage’s colourisation app is powered by the impressive DeOldify model, and delivers by far the most consistent results of all of the services we tested. On the whole, colours were realistic and sensibly chosen - even allowing for its bias towards blue in clothing - but the tool did tend to render sepia photos with a yellowish tinge. However, converting these to black and white prior to uploading them to MyHeritage eliminated the problem.
Colourise uses an updated version of the Colorful Image Colorization model, and results are largely realistic, if a little dull and lacking in vibrancy at times. Ancestry’s recently added colourisation feature is accessed through your online tree or via the mobile app - navigate to the ‘Media Gallery’ and tap an image. It’s powered by the dated Photomyne engine, which produced by far the least convincing results across the test. This included frequently introducing splotches of purple and red to clothes.
Beyond the core colourization, many services advertise enhancements such as upscaling, restoration of scratches, and background removal to create more polished final images. The following sections explore the core capabilities, strengths, and caveats of leading AI colourizers, drawing on examples and descriptions from the tested tools.
Core technologies and how they affect results
- DeOldify-based colourisation (used by MyHeritage) tends to deliver realistic skin tones and coherent clothing colours, with occasional bias toward blue in garments. It excels at preserving natural textures and details, producing consistently convincing colour renditions across varied historical imagery.
- Colorful Image Colorization variant (used by Colourise) yields largely believable colours, though images can appear somewhat dull or less vibrant, especially in areas with subtle tonal variation. The results are generally stable and suitable for many archival needs.
- Photomyne engine (used by Ancestry) is described as dated in the test, with less convincing colourisation results and occasional colour artefacts such as purple or red splotches on garments.
Some tools offer additional features beyond colourisation. For example, MyHeritage enables users to tweak the effect via multiple options and to switch the underlying model version. Other platforms provide built-in upscaling, scratch repair, and background removal to help prepare final images for printing or sharing.
Practical guidance: choosing a service for your needs
- For realism and consistency: Prefer DeOldify-based solutions (as in MyHeritage). They deliver the most consistent and natural-looking colours, with strong skin-tone reconstruction and reliable texture rendering.
- For speed and simplicity: Colourise offers quick results with a straightforward workflow and a stable, realistic palette, albeit with occasional dullness in vibrancy.
- For archival accuracy with caveats: Ancestry’s implementation (Photomyne-based) may be less convincing overall and can introduce colour artefacts, so use tools like upscaling and restoration with caution if you rely on precise colour fidelity.
In practice, a common workflow is to upload a grayscale or sepia image, wait for the AI to process, and then optionally adjust the result or upscale the image for printing. Where available, you can also convert sepia back to color by re-processing after pre-adjusting the input (e.g., converting sepia photos to black-and-white first). Some platforms also offer direct navigation to image galleries and settings to fine-tune outcomes.
Enhancements that help elevate colourized photos
- AI upscaling to increase resolution and sharpen details, making colourized outputs suitable for large prints.
- Scratch and noise handling to reduce visible artefacts and produce cleaner images.
- Background removal to place subjects in new contexts without distracting borders.
- Color consistency across clothing, fabrics, and landscapes by leveraging texture-aware models trained on diverse historical materials.
These enhancements can be particularly valuable when restoring family memories or preparing historical photographs for exhibition. It’s important to test multiple tools on a given image, as subtle differences in model biases and training data can yield noticeably different colour results.
Representative use cases and outcomes
Colorising eight historical photos across the 20th century demonstrated both the promise and the pitfalls of consumer AI colourisers. In practical terms, users should expect:
- Natural skin tones that avoid the artificial “orange” or “grey” look when the model is well-calibrated to human complexion variations.
- Accurate handling of textures like wood grain, fabrics, and metallic surfaces, depending on the model’s understanding of historical materials.
- Possible colour biases or artefacts in certain contexts, especially with earlier, more degraded images or sepia-toned originals.
Ultimately, the choice of tool may depend on the desired balance between realism, vibrancy, and processing convenience. For historians and archivists, stability and reproducibility across batches are valuable; for family memories, lifelike skin tones and faithful fabrics often take precedence.
Table: comparative highlights from tested tools
| Tool | Strength | Weakness | Best use case |
|---|---|---|---|
| MyHeritage | Consistent, realistic colours; DeOldify-based | Sepia can pull towards yellow; requires pre-processing for sepia | Portraits with diverse skin tones |
| Colourise.com | Largely realistic results; straightforward workflow | Colours can be a bit dull; vibrancy varies | Quick colourisation of mixed-era photos |
| Ancestry | Direct integration with online tree; convenient sharing | Least convincing results; artefacts like purple/red patches | Casual recolouring in family history projects |
For those seeking additional improvements, several services advertise AI-powered upscaling and restoration as complementary tools. These features can be used after the initial colourisation to produce print-ready images with higher fidelity and sharper detail.
Practical steps to colourize a photo online
Step 1: Drag and drop or upload the vintage black-and-white photo you want to colorize. Step 2: Let AI automatically analyze textures, apply historically accurate colors, and enhance the resolution instantly with the AI colorizer. Step 3: Download your vivid, restored color image in HD quality to share online or print with confidence.
While some services advertise a fully automated experience, it can still be beneficial to review the result and experiment with different model versions or adjustment options where available. This helps ensure the final image aligns with the historical context and the photographer’s original intent as closely as possible.
Colorizing photos online is now accessible to a broad audience, with multiple tools offering free access or trials and browser-based workflows that bypass complex software installations. For many users, this makes the rediscovery of old memories faster, more affordable, and more satisfying than ever before.
