Using exact colors in AI image and design prompts

By the 000098 editorial team · Updated October 4, 2026 · 6 min read

AI image generators are good at mood and style but often loose with exact colors. If you need on-brand output, a little structure in your prompt and a quick correction step go a long way.

Describe color in words and numbers

Most image models were trained on captions, which use color names far more often than hex codes. Combine both: "deep ink-navy blue (#000098) background, warm gold accents (#D9D02A)". The name steers the model; the hex code helps tools that do parse it, and documents your intent.

Give a palette, not just a color

State the proportions: "mostly off-white background, deep navy typography and shapes, small gold highlights". This mirrors the 60-30-10 rule and reduces the chance the model floods the image with your accent color.

Constrain what you do not want

Many tools support negative prompts or exclusions. Excluding "purple tint, neon, rainbow" helps keep a navy brand from drifting toward violet, a common failure with saturated blues.

Use reference images

Where supported, upload a swatch image or a previous on-brand asset as a style reference. A simple image of your palette stripes, exported from the palette generator, works surprisingly well.

Correct colors afterwards

1. Pick the generated colors with the image color picker.
2. Compare them with your brand values using the contrast checker or the color pages.
3. Use a selective color or hue/saturation adjustment in an editor to shift them back on brand.

AI design and code tools

Tools that generate layouts or code respond best to explicit design tokens. Paste a CSS variable block (each color page has one) and instruct the tool to use only those variables. That keeps generated components on brand and easy to retheme later.

Rights and disclosure

Check the terms of the AI tool you use for commercial rights, and follow platform rules on labelling AI-generated images where required.