OpenPalette launches 70B multimodal model for UI prototyping

AI · 5 min read

OpenPalette today shipped a 70 billion-parameter multimodal model trained specifically on UI screenshots, vector designs, and interaction metadata. The model ingests raster sketches, hand-drawn flows, and text prompts to generate editable vector layouts and component libraries tailored to a project's design system.

The company also introduced a plugin API that connects the model to Figma, Framer, and a new open-source editor, enabling one-click conversion of model outputs into native components. Outputs include accessibility metadata, spacing tokens, and suggested responsive breakpoints.

Early adopters report significant time savings in concepting and handoff, but some raise concerns about licensing for training assets and fidelity for complex interaction states. OpenPalette says it will publish auditing tools and a model card documenting datasets and evaluation metrics.