Runway, one of the first AI video generation platforms to publicly launch, has announced the third generation of its model— and a huge step forward for its technology

Just as OpenAI makes its end goal, for Runway, which is a common world model, says it is artificial general intelligence. It is an AI system that can build an internal representation of an environment and use it to simulate events within that environment.

Gen-3Alpha, a new model of the runway, is the closest the startup has come to achieving its long-term ambitions. The company says it will power all image and text-to-video tools on the runway platform as well as other features like motion brushes and text-to-image. 

The runway does not state when Gen-2 will be implemented, replacing the current Gen-3 model, but new safeguards have been added for Gen-3, including improved visual moderation and a C2PA standard that makes it easier to track the origin of different types of media. It's the latest in a new generation of AI video models, including OpenAI Sora, Luma Labs Dream Machine and Kling, with longer clips and improved motion.

The runway says the Gen-3 is the first in a series of models trained on the new infrastructure. This is specifically built for large-scale multimodal training and improves loyalty, consistency and movement. 

One of the lessons learned from Sora 1 is that the scale is more important than most others, so adding calculations and data can greatly improve the model.

The new model is trained simultaneously in video and image, and the runway says it improves visual quality from tex-to-video prompts.

The new model will also power new tools that provide finer control over things like structure, style and movement.

I haven't had a chance to try the Gen-3 myself and it's still in alpha mode, but the video seems to show a significant improvement in movement and quick adherence.

Each video is approximately 10 seconds long, which is approximately 2 times the default brightness length, similar in length to Sora video. It's also almost 2 times longer than the current runway Gen-3 video.

Prompt: "A subtle reflection of a woman on the window of a train traveling at ultra-high speed in a Japanese city."

Prompt: "An astronaut running down an alley in Rio de Janeiro.

Prompt: "FPV is flying through the colorful coral-lined streets of an underwater suburb.

Prompt: "Handheld tracking was taken at night, followed by dirty blue balloons floating on the ground to abandon the old European streets."

Prompt: "Extreme close-up shot of an ant emerging from its nest." The camera pulls back and reveals the neighborhood across the hill.

Prompt: "A zoom-in shot in the face of a young woman sitting on a bench in the middle of an empty school gym."

Prompt: "FPV drone shot a castle on a cliff."