DeepMind unveils Muse, a creative multimodal model for asset generation

AI · 5 min read

Muse focuses on generating production-ready creative assets rather than solely photorealistic images. Designers can request an icon set, layout system, and motion specs in one prompt; Muse outputs SVGs, variants optimized for different densities, and Lottie-compatible motion snippets alongside implementation notes.

The model is trained with a pipeline that separates structural constraints (grids, alignments) from stylistic variables, enabling it to produce consistent sets that scale across product systems. DeepMind is also releasing an evaluation suite for layout cohesion, color contrast, and downscaling fidelity.

DeepMind plans to make Muse available via enterprise API and will partner with select design platforms to embed asset packs into design libraries. They emphasized licensing and provenance features to ensure generated assets can be traced and attributed when needed.