GFX-201b · Module 1
Building Palette Prompts
4 min read
A brand palette is not a list of hex values. It is a system of color relationships — primary, secondary, accent, neutral, and the rules for how they interact. When you translate a brand palette into AI prompt language, you are encoding those relationships, not just the individual colors. The prompt needs to describe the palette as a whole: "muted earth tones with a single warm accent" communicates a complete color system in eight words. "Brown, beige, terracotta, cream, and orange" communicates five disconnected colors that the model may or may not combine harmoniously.
Do This
- Describe palette relationships: "monochromatic blues with a single amber accent"
- Reference real-world palette anchors: "the color palette of a Wes Anderson film"
- Specify the dominant-to-accent ratio: "primarily neutral with 10% saturated accent color"
Avoid This
- List individual colors without describing how they relate to each other
- Use hex values in prompts — models do not understand hex codes
- Describe more than four colors — the model will prioritize some and ignore others unpredictably
The most reliable palette technique is the reference-plus-override approach. Start with a strong palette reference — a film, a decade, a design movement — and then override the specific elements that diverge from your brand. "The warm neutral palette of a 1970s Terrence Malick film, but replace the golden tones with cool silver" gives the model a complete palette system to start from and a surgical adjustment to make. This is more reliable than building the palette from scratch in every prompt because the model already knows what a Malick palette looks like.
Document your palette prompts in your brand system as reusable fragments. When your brand palette is "muted warm neutrals with desaturated sage green accents, reminiscent of a Japanese ceramic workshop," write that phrase once and paste it into every prompt. The phrase becomes your palette token — a portable color instruction that produces consistent results across different subjects, environments, and tools.