A new AI achieves top scores turning images, text, and drawings into CAD designs.

Designing industrial parts with computers is hard. Builders want software to turn text, drawings, rendered images, or real photographs into working design code. They also need tools that assemble separate parts by figuring out how they fit together and balance. Current tools are either too narrow to handle all these inputs well or too general to stay reliable on specialized design work. To solve this, researchers built VisCAD. At its center is VisCAD-M1, a large AI model trained specifically to produce code for individual parts. It beats other evaluated models on test benchmarks. It can even check its own work to boost scores further. VisCAD also includes a dedicated tool that uses top models to assemble multi-part objects, outperforming general setups in both visual results and scored tests.

Why it matters

Industrial designers often struggle to convert diverse ideas into precise CAD models. This AI can take various inputs and generate accurate designs, potentially speeding up product development and reducing manual effort.

Questions answered