Developer projects

Run an action-conditioned interactive world model with open-ended rollout on a single desktop GPU.

Generate continuing interactive worlds with bounded external state, mid-rollout prompts and camera control.

A camera-controllable streaming video world model that grows a 360° scene prior from a single perspective image.

Build long-horizon playable video worlds with camera control, prompt switching and explicit spatial memory.

Train few-step autoregressive video generators with frame-wise and chunk-wise Causal Forcing variants.

A world model that reasons through motion and geometry with an explicit chain of thought before predicting future video.

Generate video progressively with a distilled causal diffusion model supporting streaming and dynamic prompts.

Run ComfyUI workflows on live video streams, with a browser interface and WebRTC transport.

Diffusion-powered visual agents for continuous LLM-connected conversations and long-form avatar video.

A source-available tool for running interactive generative-video pipelines, with a non-commercial alpha licence.

A hosted real-time video-to-video editing model with live prompts, references and SDK sessions.

A keyboard-controlled open-world research experience whose visuals and mechanics are generated by a world model.

A general-purpose action-controlled video world model with open-ended event response and long-rollout training.

An open-source workspace for interactive video generation and editing, with a browser interface and a self-hosted FastVideo backend.

Train playable action-conditioned video world models with an open progressive causal recipe.

Generate egocentric video worlds controlled by tracked hand poses and camera motion from VR equipment.

Long-horizon video world model rollouts that stay geometrically consistent through a fixed-size geometry-aware implicit memory.

Run a released 14B long-video model with training code, inference tools and distilled checkpoints.

A preview audio-visual world model that responds to navigation actions while generating synchronized media.

Generate multi-view-consistent frames from a reference image and target camera poses on consumer-oriented hardware.

Two released research systems in one repository: a long-horizon multi-shot audio-video generator and an interactive omnimodal world model.

Real-time instruction-guided editing of open-ended video streams with an autoregressive diffusion transformer.

Explore a controllable video world model with released base and fast inference variants.

Edit a streaming video with diffusion while reusing unchanged regions and extending to longer sequences.