Controllable Video Generation

Controllable video generation is a category of AI video technology that gives users precise control over what appears in the generated video. Instead of typing a text prompt and getting a random result, you provide specific inputs: a reference motion video, a character image, camera direction, or scene parameters, and the AI generates video that follows your instructions.

This matters because standard text-to-video tools (like Sora, Runway, or Pika) generate impressive results, but you can't reliably direct what happens. Ask for "a person dancing" and you get a person dancing, but not your person, not your dance, and not the specific motion you had in mind.

Controllable video generation solves this by accepting additional inputs beyond text. The most powerful form combines a character image with a motion reference video: you show the AI who should move and how they should move, and it generates a video where that character performs that exact motion.

Viggle's JST-1 model is a controllable video generation system. It's built on a video-3D foundation model that understands physics and 3D body mechanics, which means it can map complex motions: ballet spins, gymnastics, martial arts, rapid dance choreography, onto any character while maintaining consistent identity across frames.