Runway's Gen-4 targets consistency in AI video
Runway says its Gen-4 video model can keep characters, objects and locations consistent across scenes from a single reference image, addressing a long-standing weakness of AI video generation.
Runway describes Gen-4 as a video generation model built for "world consistency." According to the company, users can generate the same characters, locations and objects across multiple scenes and then re-render them from different perspectives and positions.
The model combines visual reference images with text instructions to produce new images and videos in a consistent style, subject and setting. Runway says this works "without the need for fine-tuning or additional training."
The page highlights several workflows:
- Consistent characters: a single reference image is used to keep a character consistent across different lighting conditions, locations and treatments. Runway says this solves "the character consistency problem" that has been AI video's core limitation.
- Consistent objects: placing a given object or subject into different environments, aimed at long-form narrative content and AI product photography.
- Coverage: users supply reference images and describe the shot composition to generate multiple angles of one scene.
- Generative VFX: output meant to sit alongside live-action, animated and VFX footage.
Runway also claims Gen-4 improves motion realism, prompt adherence and "world understanding," and calls it "a significant milestone" in simulating real-world physics and "a step towards Universal Generative Models."
The page is a product announcement. It includes no benchmarks, evaluation methods or comparisons with other models to support its claims about consistency, quality or physics. It also gives no details on architecture, pricing or access tiers. Gen-4 came out in early 2025, and Runway has since released newer versions.