OpenAI launches Muse — a design-first multimodal model for rapid prototyping

AI · 6 min read

OpenAI today announced Muse, a new multimodal model positioned as a design-first assistant that understands visual context, component hierarchies, and style systems. Muse can ingest screenshots, Figma files, and brand tokens, then propose alternate layouts, suggest copy variants, or generate assets that fit predefined constraints. The company says Muse was trained on a mix of design datasets and human-in-the-loop feedback from design teams.

Muse introduces a workflow API that ties model outputs back into design files with delta-aware edits rather than full-file rewrites. Designers can request changes such as "tighten spacing between CTA and hero copy" or "generate three mobile-first nav alternatives" and get editable outputs that map to existing components and tokens. OpenAI emphasized that the feature aims to reduce manual reconciliation work between AI suggestions and design systems.

Privacy and control were front-and-center: Muse offers project-level style locks, a clarity slider for how "conservative" vs. "inventive" outputs should be, and an enterprise on-prem option for teams that cannot send assets to the cloud. Early partners include a mix of product teams and digital agencies, who reported iteration time cut by up to 40% in pilot programs.