MIT Researchers Combine Generative AI and Physics to Create Real-World Designs

PhysiOpt system helps turn creative 3D models into structurally sound physical objects

Feb. 26, 2026 at 6:44am

Got story updates? Submit your updates here. ›

Researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a system called PhysiOpt that combines generative AI models with physics simulations to create 3D designs that are not only creative, but also structurally sound and manufacturable in the real world. The system allows users to input a prompt or upload an image, and it will generate a 3D model that can be 3D printed while ensuring the design is functional and can withstand real-world forces and constraints.

Why it matters

Generative AI models can produce highly imaginative 3D designs, but those designs often lack the structural integrity to be fabricated into physical objects. PhysiOpt addresses this gap by integrating physics simulations that test the viability of the designs, enabling the creation of unique personal items and decorations that can actually be produced and used.

The details

The PhysiOpt system works by taking the output of a generative AI model and running a finite element analysis to stress test the 3D design. It identifies areas of the model that are not well-supported and iteratively optimizes the design to make it structurally sound, all while preserving the overall appearance and functionality. Users can specify parameters like the intended use, materials, and force/weight requirements, and PhysiOpt will generate a 3D model that meets those constraints. The system uses pre-trained shape priors to efficiently create designs in various styles, like the steampunk-inspired keyholder and giraffe-shaped table created by the researchers.

  • The PhysiOpt system was presented by the researchers at the ACM SIGGRAPH Conference and Exhibition in Asia in December 2026.

The players

Xiao Sean Zhan

MIT electrical engineering and computer science (EECS) PhD student and CSAIL researcher, co-lead author on the paper presenting PhysiOpt.

Clément Jambon

MIT EECS PhD student and CSAIL researcher, co-lead author on the paper presenting PhysiOpt.

Kenney Ng

MIT-IBM Watson AI Lab Principal Research Scientist, co-author on the paper.

Mina Konaković Luković

Assistant Professor at CSAIL and principal investigator, co-author on the paper.

MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL)

The research lab where the PhysiOpt system was developed.

Got photos? Submit your photos here. ›

What they’re saying

“PhysiOpt combines GenAI and physically-based shape optimization, helping virtually anyone generate the designs they want for unique accessories and decorations. It's an automatic system that allows you to make the shape physically manufacturable, given some constraints. PhysiOpt can iterate on its creations as often as you'd like, without any extra training.”

— Xiao Sean Zhan, MIT EECS PhD student and CSAIL researcher

“Existing systems often need lots of additional training to have a semantic understanding of what you want to see. But we use a model with that feel for what you want to create already baked in, so PhysiOpt is training-free.”

— Clément Jambon, MIT EECS PhD student and CSAIL researcher

What’s next

The researchers plan to further improve PhysiOpt by incorporating vision language models to enable more autonomous, common-sense design generation, as well as modeling more complex fabrication constraints.

The takeaway

PhysiOpt represents a significant advancement in bridging the gap between creative digital designs and physically realizable objects, empowering users to bring their imaginative ideas to life in the real world through a combination of generative AI and physics-based optimization.