Black Forest Labs launches Flux 3 multimodal model for video, audio, and action

Black Forest Labs has officially announced Flux 3, its latest frontier model designed to integrate multiple data types into a single system. Unlike traditional approaches that connect separate image, video, and audio models behind the scenes, Flux 3 was jointly trained across these modalities from the ground up.

A Unified Approach to Visual Intelligence

The central idea behind Flux 3 is that a robust artificial intelligence model must learn a comprehensive representation of the physical world rather than studying individual components in isolation. Images capture spatial arrangements at a single moment, while videos introduce time, motion, and physical laws. Audio adds causal relationships between mechanical events and acoustics that visual data alone cannot detect. By learning from all of these modalities simultaneously, the system uses their mutual constraints to better understand how objects interact.

Capabilities and Early Access

Flux 3 Video is now available through an early access program, allowing users to generate clips up to 20 seconds long with synchronized audio from a single prompt. Key capabilities include text-to-video generation, image-to-video animation, video-to-video style transfer, and keyframe-to-video generation for smooth transitions. The model is also designed to handle multi-shot chaining, where individual clips are linked together to maintain character consistency across longer sequences.