Video Generation Research Intern: Controllable AI

Hoyoverse; Mountain View (CA); On-site; USD 69,000 - 110,000; Part time

Job summary

HoYoverse is asking for a Video Generation Research Intern to work with research and engineering teams on post-training methods for large-scale video generation models. The role emphasizes controllability, generation quality, and temporal consistency, with a focus on cutting-edge generative techniques.

The position requires pursuing a Master’s or PhD in a related field, strong Python/PyTorch skills, and the ability to independently implement ideas and run experiments.

Qualifications

• Pursuing a Master's or PhD in CS/AI/CV/ML.
• Strong research background with hands-on video generation experience.
• Solid understanding of modern generative modeling techniques (diffusion, flow matching, DiTs, autoregressive).
• Strong Python and PyTorch skills.
• Ability to independently implement research ideas and run experiments.

Responsibilities

• Research, develop, and implement post-training methods for video generation models, including supervised fine-tuning, preference optimization, and reinforcement learning-based approaches.
• Explore controllable video generation techniques to improve motion control, camera movement, trajectory following, and action-conditioned generation.
• Design experiments to improve video generation quality, temporal consistency, motion realism, and alignment with user instructions.
• Collaborate with researchers and engineers to develop training pipelines, evaluation benchmarks, and scalable experimentation workflows.
• Stay up to date with the latest advances in video generation and generative AI, and contribute to research publications and practical applications.

Skills

Python

PyTorch

Video generation

Research

Education

Master's or PhD in CS/AI/CV/ML

Tools

Git

Job description

HoYoverse is asking for a Video Generation Research Intern to work with research and engineering teams on post-training methods for large-scale video generation models. The role emphasizes controllability, generation quality, and temporal consistency, with a focus on cutting-edge generative techniques.

The position requires pursuing a Master’s or PhD in a related field, strong Python/PyTorch skills, and the ability to independently implement ideas and run experiments.