NVIDIA details Cosmos 3 camera-control post-training recipe
Camera-controllable world models turn static images into navigable scenes, which matters for robotics simulation, virtual production and synthetic data generation. Builders working with NVIDIA's Cosmos stack can read the post-training recipe directly and replicate the camera-control fine-tuning. The report offers a concrete reference point against other controllable video generation approaches like camera-path conditioned diffusion models.
NVIDIA published a technical report on training Cosmos 3 to follow camera controls. From a single input image, the world model generates new views as the camera tracks a subject, such as a dog, or rises and moves sideways. The report describes the post-training recipe used to achieve this controllability. The material was shared via the NVIDIA AI account on X on the platform alongside a link to research.nvidia.com.
How do you train a world model to follow camera controls? From a single image, Cosmos 3 generates views as the camera follows the dog, rises or moves sideways. Read the technical report for the post-training recipe → research.nvidia.com/labs/cosmos-la… 💬 1 🔄 1 ❤️ 3 👀 2469 📊 2 ⚡
