Generating Colour Palettes Thanks to Microgpt (fungiboletus.github.io)

A new project using MicroGPT has successfully automated the generation of unique color palettes for applications like customizable watch faces, significantly benefiting designers and developers in the AI/ML community. With the demand for millions of unique color combinations—particularly for watch faces needing a new palette every minute—the micro AI model offers a creative solution by leveraging a dataset of over 100,000 human-designed color palettes, primarily sourced from music album covers. This approach enhances variety and originality, a challenge in traditional random color generation methods. Transitioning from basic methods like sRGB to more perceptually accurate color spaces like OkLab, the project implemented advanced techniques, including Sobel filtering for image analysis to create palettes that prioritize visually interesting colors. The introduction of MicroGPT, a lightweight language model, has shown promise in understanding and generating complex color combinations, overcoming limitations faced by earlier models. While the model may produce some invalid palettes, improvements in sampling methods have enhanced its ability to create aesthetically pleasing and applicable combinations, pushing the boundaries of automated color design in AI applications.