Runway Characters transforms any single reference image — from photorealistic humans to cartoon mascots — into real-time conversational video agents that respond with synchronized facial animation.

The system generates audio-synchronized lip-sync, gaze dynamics, head movement and secondary motion at 24fps in HD resolution. Server-side response time from when a user stops speaking to when the character begins responding measures 1.75 seconds.

The model processes frames autoregressively with an effective 37 milliseconds of computation time per frame. Runway developed the technology as part of its broader research into general-purpose multimodal simulators.

Building General World Models

The Characters release represents one component of Runway's General World Models research program. The company positions video as the primary input and output modality for next-generation computing systems, supplemented by text and audio.

Runway's research portfolio includes GWM-1, described as a state-of-the-art General World Model designed to interact with the real world. The company also released Gen-4.5, which it calls the world's best video model for motion quality and visual fidelity.

The research team published additional papers on autoregressive-to-diffusion vision language models and dual-process image generation. The autoregressive-to-diffusion approach adapts existing models for parallel diffusion decoding without training from scratch.

Research Applications

Runway's dual-process distillation scheme allows image generators to learn new tasks from vision-language models through in-context learning. The framework enables multimodal controls for properties including color palette, line weight, horizon position and relative depth.

The company operates RNA Sessions, a series of talks on frontier AI and art research. Recent topics covered real-time video generation's impact on online interaction and robot policy evaluation using general world models.

Runway maintains separate product lines for creative applications, developer tools, and robotics policy evaluation. The company offers enterprise licensing for its video models and provides API access through its developer portal.