Generative world foundation models for data-driven simulation of physical AI systems.

Open-source video foundation model training framework, providing accelerated video dataset curation, multimodal dataloading, and parallelized video diffusion model training and inference.

Iterative computation with diffusion models offers a powerful paradigm for not only generation but also visual perception tasks. We unify tasks such as depth estimation, optical flow, and segmentation under image-to-image translation, and show how diffusion models benefit from scaling training and test-time compute for these perception tasks.

We explore diffusion and flow matching models under the theoretical framework of generator matching. Our analysis offers a fresh perspective on the relationships between these state-of-the-art generative modeling paradigms and how to build new generative Markov processes that benefit from both approaches.

We introduce Storytelling With Action Guidance (SWAG), a novel approach to storytelling with LLMs. Our approach reduces story writing to a search problem through a two-model feedback loop. SWAG can optimize open-sourced LLMs to substantially outperform previous end-to-end story generation techniques leveraging closed-source models.

We present a self-supervised test-time training approach for fine-tuning image superresolution models to adapt to new test distributions on-the-fly.