Not long ago, even the most advanced neural networks struggled with artifacts: a character’s face could “drift,” clothing changed on its own, and backgrounds flickered between frames. That is now a thing of the past.
Runway References allows for a character’s appearance to be locked throughout the entire video. It is enough to upload a reference in the prompt — and the character remains recognizable from the first to the last frame, with no chaos, no sudden mutations.
And that is not all. In Coverage mode, the model can automatically generate various angles from a single image. This dramatically simplifies the creation of videos where the character must be shown from multiple sides — dynamically, cinematically, and stably.
All of these features are already available in the Gen-4 model. Yes, this is the quality everyone has been waiting for. Neural networks now truly understand what visual continuity means.
