Revolutionizing AI Video Creation: Moonvalley Launches Marey, Empowering Filmmakers with Real-World Physics

The startup Moonvalley has launched its video generation model named Marey, which comprehends the physical characteristics of the real world.

The neural network was initially made available for beta testing in March. Now, the developers have rolled out a subscription plan: the basic option is priced at $14.99 per month, with additional plans available at $34.99 and $149.99.

Marey enables users to create videos lasting up to five seconds.

The company’s founders, who previously worked at DeepMind on a video generation model, state that their new product has been trained entirely on licensed data in an open manner. This approach aligns with the interests of the startup’s clientele—filmmakers eager to avoid potential legal disputes related to copyrighted AI-generated content.

According to the developers’ statement, Marey takes into account the physical properties of the real world and replicates movements based on natural laws. They plan to enhance this capability further to introduce more interactive elements into storytelling.

For instance, a video featuring a running bison across the plains can be transformed into a clip showcasing a speeding Cadillac in the same environment, with the grass and ground responding realistically to the vehicle’s movement. Alternatively, the model can superimpose an image of George Washington onto an actor, accurately capturing facial expressions, muscle movements in the forearms, and gestures akin to the original.

Among other features, Marey supports free camera movement, allowing users to alter the filming trajectory using a mouse. The model also has the capability to change the background in videos, providing directors with the ability to use original footage as a foundation for constructing the desired scene.

In the coming months, additional functionalities are expected to be introduced.

The public release of Marey places it in direct competition with the increasing number of AI video generation models, including Runway Gen-3, Luma Dream Machine, Pika, and Haiper.

Additionally, it’s worth noting that in May, Google launched the Veo 3 video generator, which is capable of producing sound as well.