Pre-Publication V2. Algorithmic models trained on existing video content that generate novel synthetic video outputs from inputs like including text prompts, still images, or existing footage. By learning complex statistical patterns of motion, lighting, and composition, they autonomously generate consecutive frames that are visually coherent and logical over time.
Deliberation Summary:
(1) General consideration regarding generative systems, where describing output as novel does not reflect how the models behave, because they can reproduce training material and insert recognizable performers into a result without a user asking for them. (2) Edits in language about coherency over time and describing more accurately what the systems do.
Pre-Publication V1. Algorithmic models trained on existing video content that generate novel synthetic video outputs from inputs like text prompts, still images, or existing footage. By learning complex statistical patterns of motion, lighting, and composition, they autonomously generate consecutive frames that are visually coherent and logical over time.