Research Scientist, Video Foundation Models

About Cantina:

Cantina Labs is a social AI company, developing a suite of advanced real-time models that push the boundaries of expression, personality, and realism. We bring characters to life, transforming how people tell stories, connect, and create. We build and power ecosystems. Cantina, our flagship social AI platform, is just the beginning.

If you're excited about the potential AI has to shape human creativity and social interactions, join us in building the future!

About the Role:

We are building a core team to develop next-generation native video and omni foundation models for multimodal generation and understanding. Our current focus is large-scale video foundation model development, spanning pre-training, continued training, and post-training for high-quality, controllable, consistent, and efficient generation. Our broader roadmap includes reference- and memory-based generation, multimodal understanding and interaction, and joint audio-video generation.

In this role, you will work on foundational research and large-scale model development across the full model lifecycle, including architecture, data, training, evaluation, post-training, training systems, and efficient inference. You will have the opportunity to shape both the technical direction and the team from an early stage.

What You’ll Do:

• Research, develop, and scale native video and multimodal foundation models, from early prototypes through large-scale pre-training, continued training, and post-training.
• Explore new model architectures, training objectives, and conditioning mechanisms for video generation, reference- and memory-based generation, multimodal interaction, and joint audio-video generation.
• Build and improve large-scale data curation, distributed training, evaluation, and post-training pipelines for high-quality and controllable generation.
• Design systematic experiments to understand model scaling, generation quality, controllability, consistency, robustness, and inference efficiency.
• Collaborate closely with researchers, engineers, and product teams to help shape the technical roadmap and, where appropriate, translate model advances into real-world capabilities.
• Contribute to research publications and open-source releases when appropriate.

What You’ll Bring:

• Strong research and engineering experience in generative modeling, including areas such as diffusion models, flow matching, DiTs, video generation, multimodal models, world models, or related fields.
• Hands-on experience training and evaluating large-scale image, video, or unified multimodal models using modern deep learning frameworks and distributed training systems.
• A strong track record of developing impactful models or systems, demonstrated through research publications, open-source contributions, production impact, or other significant technical work.
• Ability to independently own ambiguous research problems, move effectively from ideas to experiments, and work well in a highly collaborative environment.
• Specialized depth in one or more areas across the foundation model lifecycle, such as model architecture, data curation, controllable generation, multimodal understanding and conditioning, post-training and reward modeling, model acceleration, inference systems, or deployment.