OpenAI launches Muse-2: a multimodal creative model tuned for designers

AI · 6 min read

OpenAI today introduced Muse-2, a multimodal model engineered specifically for designers and creative teams. Unlike prior general-purpose models, Muse-2 is fine-tuned to preserve layer structure, color profiles, and editability when generating or transforming images from prompts and sketches.

The model produces layered outputs compatible with major design formats (SVG, PSD, Figma JSON) and supports bidirectional workflows: designers can sketch rough layouts and Muse-2 will propose editable vector treatments, or designers can request style variations that maintain component boundaries. OpenAI is also shipping a lightweight API flag that prioritizes editability over photorealism for downstream tooling.

Early partners include plugin makers and enterprise design platforms testing Muse-2’s ability to reduce iteration time across exploratory comps, icon generation, and rapid prototyping. OpenAI says Muse-2 will be available via their API with tiered compute options and an on-premise offering for enterprise customers concerned about IP and data residency.