Luyang Zhu

AI Research Scientist
Meta TBD Labs
Email CV Google Scholar GitHub LinkedIn

I am a AI Research Scientist at Meta TBD Labs, where I work on multimodal interaction models. Previously, I was a founding member at World Labs, leading its research efforts on World Models and developing its large-scale pretraining infra.

I received my Ph.D. at the University of Washington in 2024, co-advised by Ira Kemelmacher-Shlizerman, Steven Seitz and Brian Curless. I was a student researcher at Google from 2022 to 2024, during which my virtual try-on projects were published and launched as products. Before that, I obtained my B.S. degree at Peking University in 2018, advised by Prof. Yizhou Wang.

Products

Muse Realtime Voice Mode

Core contributor to the post-training of realtime interaction models.

[product launch]

World Labs

Marble

Creator of diffusion models featured in Marble, World Labs' first product.

[product]

Google

GenAI for Virtual Try-On

First author of TryOnDiffusion (CVPR 2023), which powered the launch of Virtual Try-On (VTO) for Shopping on Search.

First author of M&M VTO (CVPR 2024). The underlying technology powered several product launches at Google, including one presented at Google I/O 2025.

[VTO launch] [VTO for dresses launch]

Publications