Researchers have introduced the Autoregressive View Model (AVM), a novel approach to scene generation that produces views as a continuous stream without requiring an explicit 3D representation. AVM leverages a unified in-context interface to condition its output on images or videos, enabling the generation of short clips along specified camera paths. The model utilizes a learned prior for extrapolation and retrieves relevant views from both inputs and its generation history to maintain an active context. Trained on approximately 8 million scenes and video clips, AVM has demonstrated adaptability across various benchmarks for camera and text control, visual quality, and consistency over long horizons. AI
IMPACT Introduces a new method for scene generation that could advance visual content creation and simulation.
RANK_REASON The cluster contains an academic paper detailing a new model and its technical specifications. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →