Runway's Gen-1 restyles existing video
Runway Research's Gen-1 model makes new footage out of existing video by applying the style and composition of an image or text prompt, with the aim of producing video without filming anything new.
Runway Research describes Gen-1 as a video-to-video model that "realistically and consistently" synthesizes new footage. It takes the structure of a source video and applies the composition and style of a reference image or text prompt. The company puts it this way: "It's like filming something new, without filming anything at all."
The research page lists five modes:
- Stylization: applies the style of an image or prompt to every frame of a video.
- Storyboard: turns mockups into fully stylized, animated renders.
- Mask: stylizes only selected regions and leaves the rest of the frame unchanged.
- Render: turns untextured 3D renders into realistic output, guided by an input image or prompt.
- Customization: lets users customize the model for higher-fidelity results.
Runway says user studies show Gen-1 results are preferred over existing methods for image-to-image and video-to-video translation. The page doesn't name the baselines, the number of participants, or the evaluation method, and it gives no quantitative scores.
The page also leaves out the model's architecture, training data, output resolution, clip length, pricing, and availability. Runway calls Gen-1 one of its "pivotal steps" toward multimodal AI systems that "will enable new forms of creativity."