YIPENG WANG

Yipeng Wang β€” Multimodal Generative Models and Computer Vision

Yipeng builds multimodal generative models that create visual content and reason about the physical world.

He is a Member of Technical Staff at Microsoft AI, building the MAI-Image series. Previously, he built Marble at World Labs, Pika 1.5 and 2.0 at Pika, and 3D Photos, 3D Videos, and Hyperscape at Meta Reality Labs.

He earned an M.S./B.S. in Computer Science and Mathematics from WashU.

EmailResumeGoogle Scholar

Yipeng Wang wearing a gray hoodie and green jacket, standing in a tall building with a camera hanging from his neck, overlooking a cityscape.

SELECTED WORK

HIGHLIGHTS

Products and models I built and shipped.

A MAI-Image-2.5 editorial composition featuring a translucent blue chair, orange fabric, and precisely rendered typography.Microsoft AI Β· 2026MAI-Image-2.5Launched at #3 in text-to-image and #2 in image editing on the Arena leaderboards.With Kevin Ma and Xiaoliang Dai.A richly detailed ancient courtyard generated as an explorable Marble world.World Labs Β· 2025MarbleTrained Marble, a multimodal world model that creates persistent 3D environments from text, images, and video.With Christoph Lassner, Justin Johnson, and Ben Mildenhall.A Pika 2.0 Scene Ingredients fashion shoot combining a studio, model, and vivid green coat.Pika Β· 2024Pika 2.0Built Pika 2.0 with Scene Ingredients, ranked top five in Artificial Analysis’ Video Arena.With Demi Guo, Chenlin Meng, and Ziqian Zhong.A surreal Pika 1.5 Pikaffect transforming a toilet into a burst of soft white clouds.Pika Β· 2024Pika 1.5Built Pika 1.5, advancing video quality and introducing the playful, reality-bending Pikaffects.With Demi Guo, Omer Bar-Tal, and Chenlin Meng.Meta avatars meeting inside a photorealistic cafΓ© reconstructed with Hyperscape.Meta Reality Labs Β· 2023–2024HyperscapeHelped build the neural 3D reconstruction stack behind explorable photorealistic spaces on Quest.With Johannes Kopf and Changil Kim.The Facebook 3D Photos interface showing a depth-enhanced photo of a dog.Meta Reality Labs Β· 2021–20233D PhotosDeveloped neural view synthesis systems that turn everyday photos into depth-rich, interactive memories.With Johannes Kopf and Suhib Alsisan.

SELECTED RESEARCH

PUBLICATIONS

Research in generative models, 3D vision, and rendering.