Researchers Created AI Tool to Edit 3D Models

InstructMesh helps users identify and fix functional design flaws in AI-generated 3D blueprints.

Updated on Oct. 1, 2026 in Artificial Intelligence

Isometric editorial illustration of a complex wireframe geometric solid representing 3D mesh architecture, rendered in matte flat colors. — Researchers from MIT, Google, and Northeastern University developed InstructMesh, a tool using GPT-4 to refine 3D models for physical fabrication. AI Illustration. Upload story photo >

Live Poll

Would you trust AI software to design and print functional objects for your home?

Researchers from MIT, Google, and Northeastern University developed InstructMesh, a new design tool that uses GPT-4 to refine AI-generated 3D models. The software enables users to edit specific parts of a blueprint to ensure objects are suitable for physical fabrication.

Why it matters

While current AI models can generate the appearance of an object, they often struggle to meet the functional requirements needed for physical use. This tool bridges the gap between 3D generation and real-world utility by applying language model reasoning to geometric design.

The system pairs Microsoft's TRELLIS platform with GPT-4 to manipulate designs within a latent space. Novice users corrected design flaws 90 percent of the time, addressing issues found in 80 percent of raw generated models.

The players

MIT

The Massachusetts Institute of Technology is a private research university that served as a lead developer for the software.

Google

Google is a global technology company that partnered with academic researchers to help develop the AI-integrated design system.

Northeastern University

Northeastern University is a private research institution that contributed to the design and development of the InstructMesh software.

The details

InstructMesh allows users to highlight segments of a 3D model and use natural language or sliders to adjust geometry, such as extruding or enlarging parts. The team used this method to successfully fabricate functional items, including a whistle, a mug, and a knee brace.

October 1, 2026: The research team published the article detailing the InstructMesh project.
November 2026: Researchers will present their work at the ACM Symposium on User Interface Software.

The Tech Race

This development represents a shift from purely generative AI toward systems capable of iterative, functional engineering. By layering semantic reasoning over the Microsoft TRELLIS 3D generative system, researchers are addressing the limitations of models that lack awareness of physical requirements.

For makers and designers, this software simplifies the process of correcting flawed 3D blueprints without requiring advanced technical skills. Future updates may introduce physics simulations and augmented reality integration to further streamline the fabrication of functional physical objects.

The takeaway

This tool demonstrates that combining large language models with 3D design software can drastically improve the reliability of automated fabrication. Designers can expect future iterations to offer more advanced features like automated physics checking to ensure parts perform correctly in the real world.