# Generative Video at the Speed of Light — Keegan McCallum, uRun

## Executive summary

The presentation shifts the focus of generative video from improving quality (e.g., Sora) to maximizing efficiency and enabling long-horizon, real-time interaction. The core argument is that the limiting factor is no longer model fidelity but the serving infrastructure required for continuous, synchronized streaming. Key technical requirements include global GPU distribution, WebRTC/ICE/TURN setup, and complex pipelines capable of integrating multiple models (e.g., video-to-video transformation) in real time.

## Key takeaways

- Shift from Quality to Efficiency: The most significant axis of improvement is efficiency and long horizon generation, allowing for continuous streaming at a fraction of the cost compared to slower, high-quality generations. For example, real-time generation can be produced at about 1/100th the cost of minutes-long clips.
- Cost and Accessibility: Generative video is becoming highly accessible: $10 can now buy roughly three hours of continuous generated video, while $50 could provide up to 15 hours of interaction.
- New Interaction Paradigms: The technology enables new use cases beyond simple content creation, such as 'magic mirrors' (transforming a webcam feed in real time) and providing visual mediums for users who do not think primarily in text.

## Technical details

- Model Architecture: Helios, the model demonstrated by uRun, is a distillation of Juan 2.1 14B parameters. The field includes world models (for long-term consistency and fine-grained camera control), avatar models, and video-to-video transformation models.
- Real-Time Serving Infrastructure: Building real-time applications requires a robust global infrastructure including globally positioned GPUs, WebRTC with ICE and TURN protocols, and continuous streaming pipelines that synchronize multiple models to the user's controls frame by frame.
- Development Tools: uRun proposes simplifying development with a React component for interactive video integration and a programmable Python runtime to build complex, asynchronous pipelines (e.g., avatar or video-to-video models). Applications can be built via a CLI or MCP server.

## Practical implications

- Developers can build complex, interactive video applications using a simplified React component and Python runtime.
- The focus for future development is shifting from platform provision to 'software factories' that facilitate agent interaction with generative models.
- Real-time steering of generations allows content creation to move beyond the current 'slot machine' approach.

## Topics

Generative AI, Video Generation, Distributed Systems, WebRTC, Machine Learning Infrastructure, uRun, Keegan McCallum

Source: https://www.youtube.com/watch?v=Xln-On3syJk