OpenAI launches SketchGPT, a multimodal model tuned for designers

AI · 5 min read

OpenAI's SketchGPT is engineered to bridge the gap between rapid ideation and production-ready assets by accepting sketch photos, verbal prompts, and style references to output editable SVGs, component-aware layouts, and interaction specs. The model emphasizes structure: it recognizes layers, groups, and common component types (buttons, cards, navigation) and emits metadata that maps to design tokens and CSS variables.

The tool includes a plugin SDK for Figma and Framer, allowing designers to push generated assets directly into existing files while preserving component hierarchies and variable references. SketchGPT also supports an API for automation pipelines — teams can batch-convert whiteboard sessions into baseline prototypes and attach change history for designers to refine.

OpenAI highlights safety and control: SketchGPT provides an honesty score for inferred assets (how confident it is about layout intent) and a “constrain to template” mode to prevent unexpected stylistic drift. Early testers report substantial time savings during early-stage ideation and fewer manual redraws for hand-sketched inputs.