Pose-Conditioned Video Generation

Pose-conditioned video generation studies how to turn motion instructions into videos of people while keeping their appearance and movement coherent. Systems commonly use 2D skeletons, 3D body models, keypoints, or masks to guide diffusion models, often alongside a reference image and temporal adapters; some also control camera motion or multiple characters. The methods are evaluated in settings such as human animation, dance, sign language, and avatars, with attention to pose accuracy, identity preservation, and frame-to-frame consistency. A central challenge is balancing precise motion and interaction control with realistic appearance, stable long-term generation, and reliable separation of body motion from camera movement.