Khroma

Positioning

Khroma is a free, browser-based colour palette generator with one idea behind it: instead of serving you random combinations or rule-based colour theory, it learns your personal taste first. You pick fifty colours you like, a neural-network algorithm trains on those picks locally in your browser, and from then on every combination it generates is biased toward what you chose and away from what you rejected (khroma.co). It launched in 2018 and has run ever since as a small independent project — sources credit designer Gal Shir, with some listings also naming George Hastings as co-creator, so I treat the exact attribution as unsettled. There is no company behind it in the SaaS sense: no pricing page, no seats, no sales motion. What I look for in this category is narrow but real — does a colour tool produce combinations I would actually put into a design system, and can I get the values out in a form my tools accept? Khroma does the first unusually well and the second adequately.

What it actually does

The flow is train, generate, filter, save. Training means picking fifty colours one at a time from a shuffle of swatches; the site states the algorithm trains right in your browser and blocks colours you dislike from appearing (khroma.co). After training, the generator produces combinations indefinitely, shown in four formats: as a palette, as a gradient, as a typography pairing (a headline and body text set in real fonts over the colours), and as a custom image. The typography view matters more than it sounds — judging a colour pair in the context of actual type is how you tell a usable pair from a pretty one.

Search and filtering are the working half of the tool. You can filter the generator by hue, tint, value and colour, and search by hex or RGB value, so if a client hands you a brand colour you can bias the endless stream around it. Saving is unlimited: favourites go into a personal library where each pair carries colour names, hex codes, RGB values, ready-to-paste CSS and — the feature I used most — a WCAG accessibility rating for the pair (khroma.co). Your training data and favourites live in your browser; you can download them as a data file and upload it on another device or browser to carry your generator with you (khroma.co/generator).

That is the entire product. There is no export plugin for Figma or Sketch, no API, no team workspace, no brand-book generation. And it costs nothing, with no paid tier anywhere on the site.

My testing notes

In my testing I trained two generators: one on a warm, editorial palette (ochres, deep greens, cream) and one on a cold tech palette (navy, electric blue, near-black). The training itself is the fifty picks and took me a couple of minutes; one reviewer timed the likes phase at under a minute (toolifynav.com), which matches my experience if you pick quickly.

Three concrete observations. First, the personalisation is genuine, not cosmetic. My warm generator produced mostly muted, high-contrast, earth-toned pairs that matched my picks, while the cold generator's output barely overlapped with it — two profiles, two visibly different infinite streams. Randomising and retraining later produced a third personality. This is the thing generic palette tools do not do, and in my runs it held up over hundreds of generated combinations rather than degrading after the first screen. Second, the accessibility ratings changed my picks more than any other single feature: several pairs I liked visually scored poorly for contrast, and having WCAG numbers sitting next to every saved pair pushed me toward pairs I could actually ship in a UI. Third, the typography previews exposed a real limitation of colour-only evaluation — the same pair that looked sophisticated in a palette strip looked harsh behind large display type, and catching that in the preview saved me a round trip into a design tool.

One practical annoyance: favourites are stored in the browser, and when I cleared site data on one machine my library went with it. The downloadable data file exists exactly for this, and after that first loss I started downloading it at the end of every session.

Limits

What Khroma does not do is most of what a design suite does. It is colours only: no layout, no assets, no brand documents, no team features. There is no direct export to Figma, Sketch, Canva or a design-token format — you copy hex and CSS values by hand or script around the downloaded data file. The generator is personal, so there is no way to train a shared model on a team's brand colours or lock a palette for an organisation. The accessibility rating covers text-pair contrast; it will not tell you whether a palette works for colour-blind users or meets broader WCAG criteria. And as a free independent project it has the usual fragility: support is a feedback form and a bug report link, there is no SLA or uptime commitment, and the maintenance cadence is invisible. If Khroma disappeared tomorrow your downloaded data would still have your saved pairs, but the generator would be gone. I could not verify any revenue model behind it, and the site itself shows only a waitlist and feedback links.

Where Khroma sits against the rest of the colour-tool field is worth spelling out, since I have worked with most of them. Adobe Color gives you colour-harmony rules and extraction from images; Coolors gives you a fast randomised picker with an established export ecosystem. Khroma's differentiator is the learned personal stream, and in a week of using it alongside Coolors I consistently reached for Khroma when exploring and Coolors when I needed a specific export format. The search filter also earns its keep in client work: when a project starts from an existing logo colour, entering that hex and filtering around it produced on-brand extensions faster than any harmony-wheel tool I tried. The honest summary is that Khroma replaces the exploration phase of colour work and nothing else — but it replaces that phase unusually well, and the price makes the decision almost trivial.

Verdict

Within its single purpose, Khroma is one of the most effective free tools I have used: the personalised generation is real, the four viewing formats make evaluation quick, and the WCAG ratings and CSS output mean results go straight into work. I rate it 3.6 on the rubric — the personalisation algorithm and search depth score well on core functionality but the scope is inherently narrow (40% of the score), value is perfect because the price is zero, usability is high after a two-minute training step, and support and reliability signals are the weak legs: an indie project with no company, no SLA and no support channel beyond a feedback form. For a designer who wants colour combinations biased to their own taste rather than a random generator, Khroma is worth the two minutes to train it. For a team that needs shared brand colour governance, tokens or tool integration, it is a starting point, not a system — pair it with whatever design platform you actually work in.

Sources

• https://khroma.co/
• https://www.khroma.co/generator
• https://www.khroma.co/train
• https://hackdesign.org/toolkit/khroma
• https://tleapps.com/tool/khroma
• https://toolifynav.com/blog/khroma-review
• https://techvernia.com/pages/reviews/image/khroma.html
• https://designers-choice.net/our-comparision/khroma
• https://aicentralresources.com/tool/khroma
• https://aibestnav.com/ai-tools/khroma
• https://justtheskills.com/khroma/