Muse Image generates and edits images agentically — it calls search and coding tools to ground facts and get details like QR codes and plots right, then refines its own output. Muse Video is built on the same foundation, with native audio. Muse Image ships across the Meta AI app, meta.ai, Instagram Stories and WhatsApp; Muse Video is in early preview.
A generative model for video. I co-led the project, which trained a diffusion transformer on video and images of varying durations, resolutions and aspect ratios.
Learning to follow image editing instructions. Given a photograph and a written instruction — "make it winter", "turn him into a cyborg" — the model performs the edit directly, in a single forward pass, with no per-image fine-tuning, masks or extra descriptions. With Aleksander Holynski and Alyosha Efros.
Research on generative models for video, before video generation was a crowded field.
Computational photography for the Pixel phone — the part of the camera that is a model rather than a lens.
