Nathaniel Cohen

PhD researcher in generative video models

Kyutai & Sorbonne Université (ISIR) · Paris, France

I work on generative models for video: diffusion-based methods, spatio-temporal representations, and the training and inference efficiency that makes them fast enough to run in real time.

About

I am a PhD researcher at Kyutai, an open-science AI lab in Paris, and at Sorbonne Université (ISIR). I work on generative models for video, with a focus on diffusion-based methods and spatio-temporal representations. My goal is to make such models controllable, scalable and computationally efficient, with particular attention to training efficiency and fast inference.

Most recently that has meant turning strong but slow bidirectional video diffusion models into causal, few-step, streaming generators, using audio-driven talking-head generation as a testbed, the setting where latency is not a detail but the whole point (AURA, BMVC 2026). Before the PhD, I worked at the Technion on zero-shot video editing with image diffusion priors (Slicedit, ICML 2024).

News

Publications

2026

static/videos/aura.mp4

AURA: AUdio-dRiven streaming Avatar

Nathaniel Cohen, Nicolas Dufour, Amélie Royer, Alasdair Newson, Patrick Pérez

BMVC 2026 Also at the ECCV 2026 workshops Gen4AVC (poster) and AVGenL (oral + poster)

2024

static/videos/slicedit.mp4

Slicedit: Zero-Shot Video Editing With Text-to-Image Diffusion Models Using Spatio-Temporal Slices

Nathaniel Cohen*, Vladimir Kulikov*, Matan Kleiner*, Inbar Huberman-Spiegelglas, Tomer Michaeli
* equal contribution

ICML 2024 Proceedings of the 41st International Conference on Machine Learning