Video Generation
Overview
Video generation models synthesize or edit sequences while aiming to keep content, motion, and scene structure coherent over time. Common approaches use diffusion or flow-matching Transformers, autoregressive models, and conditioning from text, reference images or videos, audio, camera paths, and 3D geometry. Researchers also use memory, physical simulation, reward-based post-training, and inference-time guidance to improve identity, controllability, and long-horizon consistency, while distillation and sparse attention reduce generation costs. A central challenge is that visual quality and prompt alignment do not guarantee physical or geometric coherence, so evaluation is expanding beyond perceptual scores to test consistency, reasoning, safety, and task outcomes.
Latest papers
365 papers65 papers in the last four weeks, up 132% on the four weeks before. 0.4% of all new papers.