Warp-as-History repurposes visual memory for camera control

This method is a clever architectural "cheat" that unlocks viewpoint manipulation by exploiting how video models maintain temporal consistency.

• –It bypasses the need for heavy external camera encoders by treating geometric warps as the model's own past frames.
• –Repurposing the existing history-conditioning mechanism allows for zero-shot control on frozen models like Helios.
• –The "one-shot" training approach (LoRA on a single video) dramatically lowers the data bar compared to traditional multi-view datasets.
• –This signals a trend toward more efficient, "native" control interfaces that work within existing backbones rather than adding modular complexity.