Photo Colorizer
Colorize a black and white photo with colour that looks like the moment it was taken: true skin tones, believable sky, period-appropriate clothing.
- True skin tones
- 3 colour looks
- JPG, PNG, WEBP
- No watermark
Free credits to start. No watermark.
A photo colorizer does something a filter cannot: it decides, pixel by pixel, what colour the world was. Black and white photographs carry no colour information at all, so colourising photos means inferring it from what is visible, what is plausible and what the era suggests. Skin gets a believable undertone, foliage reads green rather than grey, walls get a temperature, and the picture stops looking like a document and starts looking like a memory.
This page runs three directions. Natural keeps the tones restrained and documentary, vivid pushes saturation for social and print, and vintage leans into the warm, slightly faded palette of early colour film. You can also hint at a known colour, such as a wedding dress or a uniform, so the model does not have to guess.
How colourisation actually works
The model has learned from millions of pairs of images: one black and white, one in colour. From that, it estimates the most likely colour distribution for each region, guided by what the object is, where the light comes from and how the tones sit next to each other. That is why a shirt gets a plausible colour rather than noise, and why faces usually come out warmer than walls.
What it cannot know is the true colour. A red car from 1952 and a green car from 1952 produce identical greys, so the model picks the statistically likely answer. That is the fundamental limit of every photo colorizer, and the reason a hint field matters when the actual colour is known: a photograph of a family in uniform should not be recoloured by probability alone.
Colourising a photo in three steps
- 1Upload the black and white photoJPG, PNG or WEBP up to 4 MB. Scans work, but clean them first: dust, scratches and creases get coloured along with everything else.
- 2Choose natural, vivid or vintageNatural for documentary and archives, vivid for social posts and prints that need impact, vintage for a period look with warm highlights and softer contrast.
- 3Add a colour hint, then compareIf you know the uniform was navy or the car was red, say so in the note. Look at skin tones first when judging the result; they are the fastest tell of an over-eager colorization.
Natural, vivid or vintage
| Look | Palette | Use it for |
|---|---|---|
| Natural | Restrained, low saturation, true-to-scene | Family archives, historical documents, anything shown as a record |
| Vivid | Higher saturation, stronger contrast | Social posts, prints, thumbnails where the image must grab attention |
| Vintage | Warm highlights, cool shadows, gently faded | Period portraits, retro branding, editorial layouts |
Skin is the hardest part to get right. If faces come out orange or grey, drop to Natural and add "keep skin tones neutral" to the note.
Getting a believable result
- Restore the scan before colourising: dust and cracks become coloured artefacts otherwise.
- Upscale afterwards, not before, so the model works on clean structure rather than interpolated noise.
- Give one or two known colours at most; long colour lists confuse the run.
- Check hands and teeth, where bad colorizers oversaturate and turn skin plastic.
- Keep the original black and white file: colourisation is an interpretation, and you may want the document version later.
Old photos: the honest order of operations
Three repairs compete for the same pixels, and the order changes the outcome. Restoration first, because scratches and fading are structural damage that colour would simply decorate. Upscaling next, so the model has more pixels to work with. Colourisation last, because it is the step that invents information, and inventing colour on top of a damaged scan is how you get a blotchy face.
If the print is badly faded, restore at the full setting; if it is mostly intact and only needs a lift, light restoration plus natural colour is usually enough. For a portrait with a single subject, that combination is the strongest result this toolkit can produce.
Where a photo colorizer is actually used
Families are the biggest case: a scanned wedding photo, a grandfather in uniform, a childhood picture that only exists in grey. Publishing and heritage work is the second, where colourised images make a century-old archive legible to a modern reader, as long as the caption is honest about the colour being an interpretation.
Commercial uses exist too. Retro brands colourise archive photography for packaging. Documentary makers test colour on stills before committing to a look. Genealogists add colour to records for family books. In every one of those cases, the value comes from the picture feeling present rather than from the colours being historically provable.
Credits, limits and privacy
Colourisation runs on a paid image model, so it needs an account, and the cost per model is shown before you run. New accounts start with free credits, and the cheapest model is a reasonable first pass on a clean scan. Uploads are capped at 4 MB and output renders at up to 1344 pixels on the longest edge with your aspect ratio preserved.
Images are sent for the duration of the request and the result is stored only so you can download it. Nothing is added to a public gallery, and no email address is collected. For a large print, colourise first, then use the free upscaler to reach 4096 pixels.
Why colourised photos so often look wrong
Three artefacts give bad colorization away immediately. Orange skin, because the model defaulted to a warm tone across every face regardless of lighting. Grey vegetation, because foliage in an old photo has low contrast and the model cannot tell leaves from stone. And colour bleeding across edges, where a red jacket tints the wall behind it because the boundary was ambiguous in the original.
All three are fixable. Skin improves when you ask for neutral tones and check the result against the hands, which are usually less retouched than the face. Vegetation improves when you hint at it. Bleeding improves when the scan is sharper, which is why restoration before colourisation is not optional for damaged prints.
| Artefact | What it looks like | Fix |
|---|---|---|
| Orange skin | Faces too warm, hands mismatched | Use Natural and note "neutral skin tones" |
| Grey foliage | Trees and grass stay colourless | Note the scene, e.g. "garden with green foliage" |
| Colour bleed | Tint spreading past object edges | Restore and sharpen the scan first |
| Flat sky | Sky is a single solid blue | Use Vivid and note "natural sky gradient" |
Historical accuracy: what to say when you publish
A colourised photograph is an interpretation, and the honest caption says so. The colours are the model's best guess from greyscale values, not evidence of what was there. For archives, museums and journalism, that distinction is professional practice: the image can be published, but the caption should state that colour was added digitally and is approximate.
For family use the standard is looser, but the same discipline helps. Keep the original file, label the colourised version clearly, and if you know a real colour, for example a uniform, a school blazer or a car, add it as a hint so the interpretation starts from a fact rather than a probability.
Colourising portraits, landscapes and documents
Portraits are the hardest and the most rewarding. Faces carry the most cultural weight, and the model has the most training data for them, so skin and hair usually land well while clothing can drift. Landscapes are easier for structure and harder for mood: skies and foliage have huge plausible ranges, so the level you pick matters more than the model.
Documents are the case to avoid. A colourised page of handwriting invites the model to invent ink colour and paper tone, which adds nothing and can mislead. For records, keep the greyscale scan and use image to text if you need the words.
Colourising a series so it looks consistent
An album of colourised photos almost always looks wrong before it looks right, because each image was colourised in a separate run with slightly different saturation and warmth. The cure is to fix one recipe and repeat it: same model, same level, same notes, and the same order of restoration, upscaling and colour.
Where the series has a common element, such as a uniform, a house or a wedding dress, add the known colour as a hint on every image. Consistency across the set matters more than getting any single frame historically perfect, because the viewer reads the album as a whole and small mismatches between neighbouring pages are what catch the eye.