Video Generation Content Understanding and Feedback Researcher

About the role

Lightspeed Studios, part of Tencent, develops games such as PUBG Mobile and focuses on next-generation gaming experiences. Its Lightspeed Tech Center is an R&D department that researches and provides technical services for engines, audio, QA, AI and more. This role sits within that center and researches how models understand generated video content and provide feedback.

Key responsibilities

• Research the model's ability to understand generated content, including semantics, objects, relationships and spatial structures.
• Implement state tracking and consistency modeling for generated videos.
• Explore unified generation-understanding model architectures.
• Research causal consistency control and physical constraints for the action-input to video-output pipeline.
• Research hybrid architectures for low-latency feedback in real-time interactive generation.
• Work with Agent/RL teams to build an end-to-end closed loop of generation, understanding, control and feedback.
• Validate controllable interactive capabilities in game scenarios and lead the establishment of evaluation systems.
• Track industry work and formulate proprietary technical roadmaps.

Required profile

• Ph.D. in AI-related fields such as video understanding, video prediction, reinforcement learning or multimodal generation and understanding.
• Deep understanding of video generation and VLM internals, familiarity with diffusion models, and research experience in interactive or controllable video generation.
• Ability to design complex model training pipelines independently, with strong engineering skills.
• Publications in video understanding, controllable generation or reinforcement learning; top-tier conference or journal papers preferred.
• Experience with game AI, simulation environments or closed-loop systems preferred.

Required skills

• Proficiency in Python and PyTorch
• Knowledge of diffusion models and VLMs