Intel / article

Building Runway Characters

Source: runway.com
Author: Runway
Date: Published May 4, 2026

Context

Runway's Characters system turns a single reference image into a conversational video character by generating and frames as audio arrives. Its May 2026 engineering account explains that the model's diffusion step and its decoder, which converts the model's internal representation into visible frames, run at the same time instead of one after the other. That overlap is how Runway says it reaches 24 frames per second in real time.

Smooth video and a quick conversational reply are separate measurements. Runway reports about 37 milliseconds of effective model time per frame, but 1.75 seconds of server-side delay between the user finishing speech and the first response frame, most of it (1,185 ms) in the voice rather than the video pipeline (567 ms). Network travel adds about 400 ms round trip in the session described. The system can therefore maintain fluid animation while a spoken exchange still includes a pause before the character replies.

Terms in this piece · Glossary

streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
transformer — The neural network architecture behind modern AI models, built on attention — letting every word directly consider every other word in parallel.

Read the source runway.com