Multimodal Video Generation Research Scientist

L&Q OASIS LIMITED; Greater London; On-site; GBP 68,000 - 91,000; Full time

Job summary

Tencent Lightspeed Studios at the Lightspeed Tech Center is seeking a researcher to design and implement a multimodal controllable video generation framework for next-gen gaming, with inputs from keyboard, controller, trajectory and voice.

The role emphasizes high-fidelity generation under sparse control signals, integration with game engines, and collaboration across Engine/Agent teams, including reinforcement learning post-training.

Qualifications

• Ph.D. in AI with a focus on Computer Vision/Deep Learning.
• Expertise in conditional generation and diffusion models.
• Project experience in camera, trajectory, or posture control.
• Understanding of game engine input/output logic and frame sync.
• Proficient in Python/PyTorch; publications in controllable video generation.

Responsibilities

• Design a multimodal controllable generation framework supporting keyboard, controller, trajectory, and voice.
• Research high-fidelity generation under sparse control signals and extend conditional control paradigms.
• Decouple and combine trajectory, camera, motion, and posture control across modules.
• Develop dedicated control modules for different game genres.
• Build video generation for multiplayer collaboration with low latency and inter-frame retention.
• Collaborate with Engine/Agent teams to standardize controllable signal interfaces.
• Drive joint optimization of controllable generation and reinforcement learning post-training.

Skills

Conditional generation

Multi-player control

Python/PyTorch

Publications in controllable video

Game AI / game development

Real-time video generation

Diffusion models

Frame synchronization concepts

Education

Ph.D. in AI with Computer Vision/Deep Learning focus

Tools

Python

PyTorch

Game engines (input/output)

Job description

Tencent Lightspeed Studios at the Lightspeed Tech Center is seeking a researcher to design and implement a multimodal controllable video generation framework for next-gen gaming, with inputs from keyboard, controller, trajectory and voice.

The role emphasizes high-fidelity generation under sparse control signals, integration with game engines, and collaboration across Engine/Agent teams, including reinforcement learning post-training.