Black Forest Labs Launches FLUX 3 With Video Generation Capabilities

Black Forest Labs (BFL) has launched FLUX 3, marking a significant evolution in its AI offerings with the ability to generate video content. This new model can create video clips lasting up to 20 seconds, complete with audio that is synchronized to the visuals. The introduction of FLUX 3 represents BFL's first foray into video generation, expanding its portfolio beyond still image generation.

FLUX 3 utilizes a multimodal approach, processing images, videos, and audio within a single framework. This integration allows the AI to predict video movements and synchronize audio elements such as dialogue and ambient noise effectively. In early trials, FLUX 3 outperformed competitors in preferences among human reviewers, achieving a preference rate of 77% over Runway Gen-4.5 and 93% over Luma Ray 3.2. The model also demonstrated competitive results against Gemini Omni and Seedance models, with a win rate of 52% in evaluations.

CEO Robin Rombach emphasized that FLUX 3 is about more than generating visuals; it also encapsulates a broader understanding of physical interactions. "A model that only learns images can only generate images," Rombach stated, suggesting that video generation fosters a deeper learning of physical principles such as weight and timing, essential for robotics applications. The FLUX-mimic initiative leverages this aspect, utilizing FLUX 3's predictive capabilities to train robots in practical environments, such as Audi's production lines.

Currently, the open-weight “Dev” version for external testing is set to launch later in 2026, while active access to video generation functionalities will remain limited to trusted partners. The ongoing developments hint at significant potential for integrating AI into various fields, particularly in automating and enhancing physical tasks.