Researchers at MIT's Computer Science and Artificial Intelligence Laboratory, Google and Northeastern University reported on Oct. 1, 2026, a software approach called InstructMesh that lets users refine AI-generated 3D designs before fabrication. It pairs Microsoft's TRELLIS 3D generator with the GPT-4 language model, allowing users to highlight parts of a design and describe desired changes. In examples, the team produced a dragon-wrapped mug, a shell-like whistle, a denim-look knee brace and a shrimp-shaped moving enclosure housing a motor.

To test usability, the team used TRELLIS to recreate popular models from Thingiverse. Nearly 80 percent of the generated models it produced were structurally flawed in some way, according to the report. Novices then identified and fixed these issues in InstructMesh about 90 percent of the time, as reviewed by an expert. Participants made items such as phone stands and vases.

The researchers said they planned to present the work at the ACM Symposium on User Interface Software and Technology in November. Lead author Faraz Faruqi, now at Google, described possible future directions, including an augmented reality platform, physics simulations and integration of a newer TRELLIS version; the report presented these as ideas, not completed work.