What is warp-as-history?

Warp-as-History is an AI research project for generating videos where you can control the camera path, for example, making the camera pan, rotate, or fly through a scene. The key contribution is that the system can be trained on a single annotated video example, rather than requiring massive datasets of camera-annotated footage. The core idea is to use a "warp" technique: given a starting frame and a desired camera trajectory, the system geometrically projects what parts of the scene should become visible as the camera moves. This warped view is then fed as historical context into a video generation model, which fills in the realistic details (lighting, texture, motion) that pure geometric projection cannot capture. This approach allows interactive control over where the camera goes in a generated video, similar to how some game engines and video simulation platforms operate. The README demonstrates a web interface where you can upload an image, write a text prompt, and click buttons to steer the camera, then generate the video clip in real time. The project is a Python research codebase intended for machine learning practitioners. It requires specific deep learning libraries and a GPU to run inference or training. Pre-trained model weights are downloaded separately from a model hosting service. The full README is longer than what was provided.