Oscar Michel

Oscar Michel is a computer vision researcher whose work spans text-guided 3D graphics, vision-language models and video world models. At the University of Chicago, he studied mathematics, received the McCormick Fellowship and coauthored Text2Mesh, a text-guided mesh-stylization paper selected for a CVPR 2022 oral. As an Allen Institute for AI predoctoral researcher, he coauthored Objaverse, first-authored OBJECT 3DIT and contributed a fine-grained world-knowledge benchmark to Molmo and PixMo. In 2025, he coauthored PISA Experiments on physical accuracy in video generation, received an NSF Graduate Research Fellowship, and Molmo and PixMo received a CVPR Best Paper Honorable Mention. He is a PhD student at NYU Courant advised by Saining Xie, and his world-model work includes the 2026 multiplayer Minecraft model Solaris.

Works: 13
Citations: 1,243
h-index: 5

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

2021

The McCormick Fellowship

As a mathematics undergraduate at the University of Chicago, he works with Rana Hanocka and Michael Maire. He receives the McCormick Fellowship in 2021.
2021–2022

Text2Mesh earns a CVPR oral

In December 2021, he coauthors Text2Mesh with Roi Bar-On, Richard Liu, Sagie Benaim and Rana Hanocka. The framework stylizes a 3D mesh to follow a text prompt, and CVPR selects the paper for an oral presentation in 2022.
2022–2023

Objaverse's annotated 3D models

As a predoctoral researcher at the Allen Institute for AI, he works mainly with Tanmay Gupta and Ani Kembhavi. In December 2022, he coauthors Objaverse, a dataset of 800K+ annotated 3D models that appears at CVPR in 2023.
2023

OBJECT 3DIT for image editing

He is first author of OBJECT 3DIT, submitted in July 2023. The paper introduces OBJECT, a dataset of 400K image-editing examples created from procedurally generated 3D scenes.
September 2024

Molmo and PixMo

In September 2024, he is one of 50 authors of Molmo and PixMo, a family of open vision-language models released with weights, datasets and source code. His contribution is a benchmark for fine-grained world knowledge in VQA.
March 2025

PISA Experiments

In March 2025, he coauthors PISA Experiments, which studies physical accuracy in large video-generation models. The paper finds that fine-tuning on simulated videos can induce object freefall and includes a benchmark for measuring physical accuracy.
2025

CVPR honor and NSF fellowship

CVPR gives Molmo and PixMo a Best Paper Honorable Mention in 2025. He receives the National Science Foundation Graduate Research Fellowship in 2025.
Since 2024

PhD research at NYU Courant

At NYU's Courant Institute, he is a PhD student advised by Saining Xie and researches world models; he joins the group in Fall 2024 with an NSF fellowship. In February 2026, he coauthors Solaris, a multiplayer video world model for Minecraft, using a data system that collects 12.64 million multiplayer frames. The paper is a NeurIPS 2026 spotlight.