Research Engineer/Scientist (all levels), World Models

About the Team The Vision-Applied Research team focuses on applied research in Generative AI and CV/Multimodal Understanding, and delivering intelligent solutions to Tiktok, enabling users to make and share creative content in a much easier way. The team has research groups dedicated to generative models for content creation, image generation, video synthesis, intelligent image/video editing, and world models.

The team is looking for Research Engineer / Scientists who can take initiatives in building next-generation World Models. The candidate will work on developing methods and infrastructure to train large-scale generative models from massive simulated and real-world multimodal datasets. This role places a particular emphasis on ensuring long-horizon temporal consistency, realistic physics, complex dynamics from the model and enabling users and agents to interact with the model in real-time.

Responsibilities - Develop large-scale, diverse, and interactive multi-modal data generation pipeline. - Develop training pipeline for long-context interactive video generation models. - Advance video generation models to capture long-horizon temporal consistency, realistic physical dynamics, object interactions, and causal relationships from large-scale multi-modal data.

Minimum Qualifications: - M.S or Ph.D. in Computer Vision, Computer Graphics, Machine Learning, or equivalent experience. - Extensive research experiences in broad GenAI, multimodal foundation models, or Embodied AI areas. - Demonstrated ability to communicate complex technical concepts and collaborate effectively within cross-functional research teams

Preferred Qualifications: - Proven experiences in at least one of the following areas: video generation and synthesis; efficient and real-time diffusion models; 3D/physics-based simulation; or reinforcement learning for agentic environment interaction. - Proven track record of first-author publications in prestigious venues including CVPR, ICLR, NeurIPS, SIGGRAPH, and ICML