OpenAI launches InstructCanvas, an instruction-tuned model for prototyping UX flows

AI · 5 min read

InstructCanvas is tuned on annotated UX datasets and designer workflows so it can take prompts like create an onboarding flow for a fitness app and return a sequence of annotated screens. Outputs include layout suggestions, component mappings, and copy placeholders to help designers iterate faster.

OpenAI built a Figma exporter that converts the model's structured output into editable nodes, preserving tokens and component references for downstream updates. The company also released a Playground for designers to experiment before integrating the API into tools or internal systems.

To address concerns about hallucinated UI patterns, OpenAI added an explainability panel that shows which training examples influenced each suggestion. The model is available via the OpenAI API with enterprise options for fine-tuning on private design systems.