OpenAI launches Fusion-2: a multimodal design model tuned for UX workflows

AI · 5 min read

OpenAI's Fusion-2 combines image, layout, and text understanding into a single model tailored to designers' needs. The model can ingest screenshots, Figma files (via plugin), and design tokens to produce annotated wireframes, accessibility audits, and suggested microcopy. It also supports explicit constraints like brand color tokens and grid systems.

A notable emphasis in Fusion-2 is privacy: OpenAI provides an option for on-prem and edge deployments that keep designs off public clouds. Enterprises can run Fusion-2 via a managed private endpoint or deploy a smaller distilled variant on local Mac hardware for quick iterations. Pricing includes a per-seat subscription with a usage tier for heavy inference workloads.

Integration partners were announced alongside the release, including Figma, Sketch-compatible plugins, and CI/CD hooks for design systems. Early adopters report faster ideation cycles and substantial time savings on mundane tasks like component naming and content placeholder generation, while some designers caution against over-reliance for intent-driven UX decisions.