Beeble launches video-to-video AI model SwitchX 3.0

Beeble has launched SwitchX 3.0, the latest version of its video-to-video AI model. 

SwitchX allows users to transform filmed environments, adding dynamic action, changing backgrounds, and directing events while keeping the original performance.

3.0 adds several new features, including support for up to four visual references, dynamic backgrounds, improved camera tracking and more precise prompt control. Users can now populate empty streets with moving crowds, transform locations during complex camera movements or direct specific actions, such as birds taking flight as an actor passes. 

The generations can be guided by reference images, floor plans and storyboards, or a combination of those. While generating new content, SwitchX uses Beeble’s propietary Source-Locked Rendering technology, which aims to preserve the original performance within an actor’s alpha matte, frame by frame, while changing the surrounding environment. 

Integrated camera tracking keeps generated environments aligned with handheld, orbiting and running shots without requiring tracking markers or a separate matchmove pass.

SwitchX 3.0 is built on MiniMax H3.0, which has been re-engineered using Beeble’s own dataset and architecture for professional video-to-video workflows. It supports up to 4K output with 10-bit colour while retaining frame count and aspect ratio from the source footage.

SwitchX is available globally through the Beeble web app and Canvas, with no change to existing credit pricing. It is included in all paid plans, with 4K and 10-bit MOV downloads available to Professional and Team subscribers.

Hoon Kim, founder and CEO of Beeble, said: “SwitchX 3.0 gives filmmakers the freedom to direct what happens around a performance, from the environment itself to the movement and timing of individual elements. Crucially, the original performance stays intact, giving artists the consistency and control they need for professional production.”