What is Palette and how does it work?
Palette.fm takes a black-and-white photograph and generates a colourised version, offering more than twenty style filters and a keyword box where you can nudge specific choices, describing the boat as red rather than letting the model guess.
Free use gives unlimited previews at reduced resolution with a watermark, plus one credit for a full-resolution, watermark-free download. Beyond that, credits are bought through a subscription or a one-time pack.
Palette standout strengths
The one genuinely free full-resolution result is the right call for this category. Colourisation quality varies a lot by photo, and letting someone test on their actual image before paying is more useful than a generic sample gallery.
Keyword control is a real differentiator against simpler colourisation tools. Being able to specify a colour rather than accept whatever the model defaults to matters a lot for family photos where someone remembers the actual colour of a dress or a car.
The pay-once option suits the most common real use case well: someone digitising a shoebox of family photos once, not colourising images every month. $49 for 75 credits with a two-year window fits that better than a forced subscription.
Palette weaknesses and drawbacks
The free tier is a preview mechanism, not a usable tool, and that should be clear going in. Watermarked, capped at 500x500, that is enough to judge whether the colours look right and not enough to use anywhere.
The economics deserve a clear-eyed look. A family archive of a few hundred photos, restored properly, runs into real money at $0.15 to $0.65 a credit. That is not unreasonable for the compute involved, and it does mean a big project costs more than the headline numbers suggest until you do the multiplication.
The core honesty point about colourisation applies here as everywhere: the model is generating plausible colour, not recovering real colour. A black-and-white photo carries no colour information to recover, so any colourisation is an informed guess constrained by whatever you tell it. Treat results as an artistic reconstruction, not a historical record, and say so if you share them with family who might otherwise assume accuracy.
Consistency also varies with source quality. Damaged, faded or low-contrast originals produce less reliable results, sometimes with colour bleeding across edges that need manual cleanup afterward.
Palette pricing & plans (2026)
| Plan |
Price |
Credits |
Cost per credit |
| Free |
$0 |
1 full-resolution credit, unlimited watermarked previews |
n/a |
| Pay-Once |
$49 |
75 credits, valid 2 years |
~$0.65 |
| Subscription |
$72/year |
480 credits/year |
~$0.15 |
Pay-Once suits someone digitising a family archive as a single project. The subscription suits someone doing this regularly, such as a genealogy service or a creator running a recurring restoration series. Neither suits someone who needs historically verified colour rather than a plausible reconstruction.
Who is Palette best for?
| User type |
Why it fits |
Considerations |
| Family archivists |
Pay-once tier fits a single big restoration project |
$49 for 75 credits, plan the batch before buying |
| Genealogy and history creators |
Subscription suits recurring restoration content |
480 credits a year at $72 is efficient at volume |
| Casual one-off users |
One free credit tests real quality first |
Everything beyond one photo needs payment |
| Historical accuracy needs |
Poor fit |
Colourisation is a plausible guess, not a record |
| Heavily damaged photos |
Mixed fit |
Results are less consistent and may need manual cleanup |
Palette review: final verdict
Palette.fm does colourisation well within the honest limits of what colourisation can be. The free full-resolution credit is a fair way to judge it, the keyword system gives real control, and the pay-once option matches how most people actually want to use a tool like this: for one project, not a subscription.
It loses marks for a preview tier that produces nothing usable, for per-credit costs that add up on a large archive, and for the underlying limitation that no colourisation tool can actually know what colour something was.
3.6 out of 5. A solid choice for restoring a specific batch of old photos. Be honest with anyone you share the results with about what the colours actually represent.
