Runway’s Gen-1 is designed to change what appears in an existing video through text prompts and other inputs. The model can turn filmed subjects into different visual forms, modify selected objects and add textures to 3D objects. Its early results can still look distorted or unreal, and the question of open-source access remains unresolved.
Transforming footage with prompts and images
Gen-1 can take filmed video and render it in a different visual style. In one example, footage of a train door is transformed into a cartoon-like train door with a text prompt. An actor can become a cartoon superhero, with an input image guiding the transformation.
Runway says users can refine the model with their own images to improve these changes. That gives the system more than a single instruction to work from: the prompt describes a desired result, while an image can help define its appearance.
The examples also extend beyond realistic footage. A few loosely assembled notebooks can become a city backdrop filled with skyscrapers through a text command. Runway describes this as “filming something new, without filming anything at all,” pointing to a workflow where a creator can build a scene from modest visual material.
Changing objects and adding surface detail
Gen-1 can isolate objects in video and alter them. Runway demonstrates a golden retriever becoming a Dalmatian, showing how the model can change a subject while working within existing footage.
The system can also apply textures to 3D objects that do not already have them. Those textures are generated from a text prompt. Taken together, these functions suggest a video editing tool that can reshape subjects, scenes and surfaces through instructions rather than relying only on conventional editing steps.
These capabilities point to a broader role for generative AI in video production. The model is being presented as a way to modify material already captured, while also creating visual elements that may not have been filmed in the first place.
Early results still have visible limitations
Gen-1 is not yet a substitute for professionally edited video. The results can contain image errors and distorted geometry, or simply appear fake and unreal. Runway frames the project as being at an early stage, so the demonstrations show possibilities rather than a finished editing experience.
The company says image and video synthesis systems are becoming more precise, realistic and controllable. If progress in image generation carries into video tools, systems such as Gen-1 could become more influential in video processing over time. That is a forward-looking possibility, not a claim that the current output already meets professional standards.
Access, openness and Runway’s plans
Whether Gen-1 will be open source is still undecided. Ian Sansavera, Runway’s video workflow architect, said the software was at “day zero.” Interested users can sign up for a waiting list, and Runway said a scientific paper would be published soon.
Runway is known for AI-powered video editing software intended to simplify and automate editing. The company showed an integration of Stable Diffusion into its toolkit in the fall of 2022. Gen-1 may primarily be developed for Runway’s own video software, though the source article describes this as likely rather than confirmed.
Runway helped launch Stable Diffusion, an open-source image AI, with Stability AI, LMU Munich, Eleuther AI, and Laion. That history makes the unresolved access question notable, but it does not settle what the company will decide for Gen-1.
Other organizations are also pursuing AI video systems. Google is working on text-to-video systems that can edit and generate video from scratch; Dreamix focuses on editing video through text prompts; and Meta has introduced a text-to-video model called Make-a-Video.