From Prompt to Colour Palettes — DataMonster
☀️ 🌙
ENG ESP
AI Dataviz Colour Tools

From Prompt to
Colour Palettes

How a recurring dataviz frustration became a proper AI-powered colour tool, and everything I learned along the way.

Pablo Gomez · DataMonster · 7 min read · May 2026

If you have spent any time building dashboards, you know the feeling. The data is ready, the layout is done, the story is clear. And then you stare at a blank colour picker for twenty minutes. Is #3A86FF different enough from #4895EF? Will that teal clash with the client’s brand? Is this palette readable for someone who is colour-blind?

Colour in data visualisation is deceptively hard. It is not just aesthetic. It determines whether your audience reads the chart correctly, whether hierarchy is clear at a glance, and whether a colourblind colleague can follow the story at all. For years I dealt with this the same way most practitioners do: bookmarked palettes, trial and error, the occasional ColorBrewer lookup. None of them understood context.

In late 2023 I decided to do something about it. What followed was a three-stage journey: a ChatGPT Custom GPT first, then a refined Claude AI skill, and finally a full web application I am genuinely proud of.

01
ChatGPT
Custom GPT
02
Claude
AI Skill
03
palette.
data.monster
Chapter 01

The GPT Experiment

ChatGPT had just launched Custom GPTs, and building a “colour advisor” felt entirely within reach. I wrote a system prompt, embedded some colour theory principles, and shared it through DataMonster channels. For a proof of concept, it worked. Describe your theme and get reasonable hex suggestions back in seconds.

But the friction was immediately obvious. Every session started from scratch. No memory, no export, no Tableau XML, and no way to preview the palette in a real chart before committing. It was a conversation about colour, not a colour tool.

The feedback I heard over and over: “This is cool, but how do I actually get it into Tableau?”

🖼 Screenshot: ChatGPT Custom GPT, the original Palette Advisor conversation interface
A useful experiment, but a conversation about colours and a usable colour tool are very different things.

The Custom GPT validated that practitioners wanted AI help with colour decisions. But it also made the gap between an interesting conversation and a usable workflow tool very clear.

Chapter 02

Levelling Up with Claude

When I started working more seriously with Claude, I rebuilt the concept from scratch as a proper structured skill. Not just a system prompt with a persona, but a carefully engineered pattern with defined inputs, controlled output formatting, and colour theory embedded directly into the reasoning.

The improvement was significant. Outputs became more consistent and context-aware. I could build the prompt to account for palette type (categorical, sequential, diverging), accessibility requirements, platform context, and the emotional register of the visualisation. The gap between “here are some colours” and “here is a considered palette for your specific context” shrank considerably.

🖼 Screenshot: The Claude AI skill, structured palette output with hex codes and export-ready formatting
Structured prompting with Claude produced far more consistent and context-aware results than the GPT experiment.

But the ceiling was clear. A skill is still a prompt. No persistence, no chart preview, no export button. The step from receiving a palette to using it in Tableau was still manual labour every single time. This needed to become a proper product.

Chapter 03

Going All In: palette.data.monster

So I built it.

palette.data.monster launched as a full web application, the project I am most proud of in the DataMonster journey so far. It takes everything learned from the GPT experiment and the Claude skill, and wraps it in an experience that genuinely respects the practitioner’s workflow from the first description all the way to the final export.

🖼 Screenshot: palette.data.monster, full app showing theme input, generated palette, lock controls and chart preview
Context-first generation, lock-and-evolve iteration, live chart preview, and one-click export to every major dataviz platform.

Here is what the tool does in practice:

🎯
Context-first generationDescribe your viz theme, style, palette type and colour count. The AI generates for your specific scenario, not a generic database palette.
🔒
Lock and evolveLock the colours you love and regenerate the rest. Iterate without losing what is already working.
📊
Chart preview firstBar, line and donut previews render with your actual palette on simulated data. Spot contrast issues before they reach the client.
📤
Export to everythingTableau XML, Power BI JSON theme, Flourish hex list, D3 array, and a Markdown brief for clients. Every format a dataviz practitioner needs.
📷
Brand colour extractionDrop in a brand guideline image or logo. The tool extracts every colour automatically, with no manual hex-eyedropper work needed.
Accessibility built inColour-blind safe mode is a first-class workflow option throughout, not an afterthought at the bottom of a settings page.

Available in English and Spanish, a nod to the ComuniDatos community where the earliest versions were first shared and where the feedback that shaped many of these features originated.

AI generates options, but humans generate meaning.

The palette generator is built entirely on this principle. It does not replace your judgement as a designer or analyst. It accelerates the process of reaching a strong starting point, so you can spend your energy on what actually matters: does this palette tell the right story for this data and this audience?

Colour in dataviz is not decoration. It is signal. It encodes categories, magnitudes, warnings and narratives. The right tools should make it easier to get that signal right.

Ready to try it?

Build your perfect dataviz palette

Free to use. Powered by Claude AI. Supports English and Spanish.
Exports directly to Tableau, Power BI, Flourish, D3 and more.

Try the Palette Generator →