NVIDIA's JCA co-generates video and 3D motion for rigs of any topology
HaggaiMaron · x · 2026-10-21
NVIDIA Research (Noam Rotstein, Yoad Tewel, Haggai Maron, et al.) unveiled Joints, Camera, Action! (JCA), a model that jointly generates video and synchronized 3D motion for rigs of any topology from a text prompt.
- Prior video-motion models either handle only human skeletons (often adding billions of parameters) or triangulate 3D motion from generated 2D pose maps; JCA keeps 3D motion in a separate model supporting arbitrary skeletons, connected to a pretrained video model via a small set of cross-attention layers
- The separation lets the motion model first learn how skeletons move from motion-only data, which is easier to obtain and augment
- The model learns to locate each joint in video frames, recovering the camera to place 3D motion in the scene
- Supports scene-aware animation, keyframe-controlled trajectories, and bidirectional video-to-motion / motion-to-video conversion
Related event: NVIDIA Research Releases JCA for Joint Video and 3D Motion Generation(2 posts)→
More from Research
- Poking into Qwen3-8B weights reveals simpler neuron activation in early and late layers — maxsloef · 2026-10-21
- How Qwen3-8B neurons work: gate/up detect concepts, down writes back to the stream — maxsloef · 2026-10-21
- Nature paper: 3-agent AI health assistant beats single-LLM chatbots — EricTopol · 2026-10-21
- CCG repeat in FAM193B promoter doubles Alzheimer's risk in Chinese adults, Nature Neuroscience study finds — anshulkundaje · 2026-10-21
- FAR.AI: activation probes catch sabotaging AI agents at a fraction of monitor cost — OwainEvans_UK · 2026-10-21
- Paper asks 'How long is the coast of a neuron?' — resolution shapes neural computation — neurovium · 2026-10-21