Hugging Face launches Palette: an open multimodal model for design prototyping

AI · 5 min read

Palette is a community‑curated multimodal model that accepts sketches, screenshots, and prompts to output layered prototypes, CSS snippets, and accessibility annotations. Hugging Face packaged the weights with example pipelines for converting Sketch and Figma files into editable outputs.

The release emphasizes transparency: model checkpoints, fine‑tuning recipes, and datasets used for layout reasoning are public. Hugging Face also launched a model card detailing limitations, including sensitivity to noisy screenshots and a documented failure mode when generating interaction states for nested components.

Developers can deploy Palette via Hugging Face Hub, run locally with accelerated inference, or use hosted inference for team trials. The community has already produced plugins for Figma and Framer that wire Palette into editor toolbars, making lightweight AI assistance accessible to indie teams.