HeadsUpAI

Runway Achieves 1.75 Second Latency for Real Time HD Video Agents

· Updated

Runway, an AI creative platform for video generation, released performance benchmarks for Runway Characters, a real-time video agent API. The system converts a single reference image into a conversational avatar streaming at 24 frames per second in HD. It achieves 1.75 seconds of end-to-end latency.
End-to-end latency
1.75 seconds
Streaming frame rate
24 fps
Resolution
HD
Effective model time
37 milliseconds per frame
Input requirement
Single reference image
Availability
Runway API

This update validates the transition from asynchronous video generation to live interaction. By reducing effective model time to 37 milliseconds per frame, Runway enables Runway's real-time interactive avatars that maintain fluid dialogue. This shift positions video as a viable interface for live digital presence and customer-facing applications.

You can deploy these agents into Runway's live video meeting integration or use the API to build custom conversational experiences. The system supports Runway Characters' real-time vision and Runway's custom voice cloning. Developers can access these capabilities through the Runway API to integrate expressive, low-latency characters into existing platforms.

View the full update on runwayml.com
Runway
Runway
@runwayml
X

Real-time video agents are here. Today, we’re sharing how we built Runway Characters, allowing you to turn one image into a fully expressive, conversational video agent streaming at 24 frames per second in HD. With just 1.75 seconds of end-to-end latency. Learn more below. https://t.co/CJqv3Kdl0v

123retweets1klikes
View on X

Still wondering? A few quick answers below.

Runway Characters is a real-time video agent API that transforms a single reference image into a fully expressive, conversational AI avatar. Unlike traditional video generation that requires rendering time, these characters are designed for live interactions, simulating facial expressions, lip-syncing, and gestures in real time to act as a conversational interface for users.

Runway Characters achieves an end-to-end latency of 1.75 seconds, measured from the moment a user stops speaking to when the AI character begins its video response. This low latency is supported by an underlying video model that processes frames in just 37 milliseconds, ensuring the conversation feels natural and responsive during live streaming.

The system streams video at 24 frames per second in high definition resolution. This frame rate ensures smooth, lifelike motion for the AI agents during conversational interactions. The performance is optimized to maintain this high-quality output consistently across the duration of a conversation without degradation in visual quality or synchronization.

Developers can access Runway Characters through the Runway API to integrate real-time video agents into their own applications, websites, or platforms. The technology is also deployable in live video meetings on services like Zoom and Google Meet, and it supports additional features like custom voice cloning and real-time vision via camera or screen sharing.

No, Runway Characters only requires a single reference image to generate a fully conversational video agent. The AI model uses that one image to build a character capable of a wide range of expressions and movements. This allows users to create stylized or photorealistic avatars without needing extensive video training data or multiple source files.

Share this update