OpenAI releases Orpheus: a multimodal model tuned for UI generation

AI · 5 min read

Orpheus mixes visual understanding with structured output, accepting screenshots, hand-drawn sketches, and text prompts to produce high-fidelity mockups plus code snippets for React, SwiftUI, and Compose. OpenAI emphasized control parameters for layout fidelity and accessibility, letting teams trade creativity for stricter spec adherence.

The model ships with a new Figma plugin that converts canvas selections into editable components and an API that returns semantic trees alongside image outputs. Early tests show the model can regenerate complex layouts and propagate design tokens, though edge cases remain around deeply nested interaction states.

OpenAI also introduced a moderation layer designed for UI contexts to minimize generation of unsafe or misleading interface elements. Pricing and enterprise deployment options prioritize hosted inference with a hybrid on-prem gateway for sensitive projects.