Haocheng Feng

Achieve fine-grained control and creative flexibility in human-environment video synthesis without heavy 3D pre-processing, thanks to a novel spatial-decoupled motion injection technique.

Instruction-guided video editing can achieve impressive zero-shot performance simply by pre-training on motion-centric video restoration tasks *before* fine-tuning on paired editing data.

Generate realistic and controllable videos of humans interacting with objects using only sparse motion cues, like wrist positions and object bounding boxes.

Achieve unified image generation by progressively disentangling and weaving together concept and localization representations within a diffusion framework, outperforming prior methods on diverse tasks.