From the archive

Runway introduces Gen-4 with reference-based video generation

Gen-4 uses visual references to help carry characters, objects, and locations across generated scenes.

By Chat Overview Published Updated

Runway introduced Gen-4, a media-generation model focused on keeping characters, objects, and locations recognizable across scenes. The model combines visual references with written instructions to guide image and video generation.

Building scenes around a reference

An image gives the model a concrete starting point for a subject or visual style. The prompt then describes the shot, composition, or movement. That approach helps with work where several clips need to look as though they belong to the same production.

Examples include:

  • Showing the same character in different settings.
  • Placing a product in several locations or compositions.
  • Exploring camera angles while retaining a scene's visual identity.

Runway presented the release as a way to improve continuity without separate model fine-tuning.

Continuity still needs editing

Reference-based generation gives more control over the starting material. Finished sequences still need a review for changes in appearance, motion, and details between shots. The time and credits spent generating alternatives are part of the production workflow.

The Runway profile covers its media tools, developer access, and separate application and API billing.

Source