Generates 5-second text- or image-conditioned videos with synchronized 44 kHz audio (including lip-sync) and built-in super-resolution to 1920×1080; available in Lite (3B) and Pro (29B) variants with code and checkpoints released under an MIT license.
Scales visual history for real-time robot control by combining autoregressive video pretraining with a streaming, asynchronous predict-then-act pipeline; shows longer context improves long-horizon manipulation and runs full inference in 107.4 ms per action chunk on an RTX 5090.
Converts static 3D Gaussian Splatting scenes into endlessly looping 3D cinemagraphs by inferring plausible dynamics with a vision-language model, synthesizing a reference video, lifting it to multi-view videos, and fitting a Fourier-parameterized Periodic Deformation Field with a Grounded Drift Field—mask-free capture of deformation, object motion, and illumination changes.
Converts pretrained multimodal LLMs into a System One decision interface that outputs structured, software-friendly decisions for video (temporal spans or multiple-choice answers); uses a shared scoring head and cached/joint inference to raise temporal-grounding mIoU and drastically speed up repeated queries.
Compiles a screenplay into shot requests annotated with explicit location, camera and event-time references to enable spatially and temporally consistent long-form agentic video generation. Uses a scene memory of registered geometry and an event memory of time-indexed observations, reconstructs and registers clips, and only admits updates after review and state checks to reduce cross-shot drift.
Trains video diffusion models to use multimodal language-model representations for generating video from sparse inputs and controlling factors such as characters, scenes, actions, and camera movement.
Generates subject-consistent videos from multiple reference images, using reconstructed surface-normal maps to guide geometry as the camera moves around the subject.
Infers implicit user intent from continuous video and adapts both response timing and content as goals change. Its benchmark covers explicit, implicit, and dynamic intents, while hierarchical memory tracks state across sustained interactions.
Provides minute-scale continuous gameplay video, camera trajectories, and structured annotations for training interactive world models that remain coherent over longer rollouts. Its two-minute sequences address the shortage of long-horizon training data.
Adapts temporal priors from video models to conditional image generation, selecting transformation scales based on each input. A latent intermediate state guides how condition information reaches the target, with evaluations spanning subject generation, edge-to-image synthesis, and editing.
Generates continuing video segments from multi-turn natural-language instructions and camera control. Its segment-wise causal attention preserves scene context while confining action instructions to their intended time windows.
Parses documents, images, audio, and video into structured JSON grounded in space and time. A shared encoder and output schema cover ten parsing tasks; the 5B model scores 96.96 on OmniDocBench v1.6.