Runway’s real-time video research changes generation into a frame-by-frame process that can stream its output. Its technical account describes adapting existing video models, then training a faster model to produce frames in fewer computational steps. The aim is to shorten the wait before video appears and support interactive experiences.
A central problem is that a visual mistake can become the starting point for subsequent mistakes. Runway therefore trains the faster model on sequences it generates itself, exposing it to the imperfect context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → it will encounter during use. This connects speed with continuity: streamingSending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.Full definition → requires both producing frames quickly and limiting the errors that accumulate between them.
Terms in this piece · Glossary
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
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.
