MIT researchers introduce GenCAD, an AI model that turns images into editable parametric CAD programs and 3D solids

GenCAD is an image-conditioned generative model for computer-aided design that produces both a 3D solid and its full CAD command sequence. The system combines transformer-based representation learning, contrastive alignment between images and CAD programs, latent diffusion, and a decoder that outputs editable parametric commands.

The work targets a key gap in AI-assisted engineering: generating designs that are not just visually plausible but also precise and modifiable. By focusing on parametric CAD rather than meshes or point clouds, GenCAD points toward AI tools that could better support manufacturing, design iteration, and engineering workflows.