Released in Nov 2025, this closed-source model uses a reference video as a strong stylistic and dynamic guide. As part of the unified O1 architecture, its input can combine a source video with text prompts, images, and even multi-angle "Elements." This allows it to generate a completely new scene that inherits the camera movement or character actions from the reference video. It prioritizes consistency over novelty, making it ideal for applying a specific visual treatment across clips or ensuring a consistent feel in a sequence. This high-control, multi-modal workflow is a key feature of the O1 family.