Controllable Video Generation Research Intern

Design and research a multimodal controllable video generation framework supporting keyboard, controller, trajectory, and voice inputs. Explore high-fidelity generation under sparse control signals, decouple and combine trajectory, camera, motion, and posture control, and build control modules for different game genres. Develop low-latency multi-player collaborative control with inter-frame state retention, define standardized interfaces with Engine/Agent teams, and optimize controllable generation with reinforcement learning post-training. Requires Python/PyTorch, diffusion models, conditional generation, and publications in controllable or conditional video generation.