Black Forest Labs releases FLUX 3: multimodal model for images, video, audio, and robot action prediction

Black Forest Labs (BFL) has released FLUX 3, a multimodal foundation model that learns from images, videos, and audio inside a single architecture. It is the first FLUX model to ship video, audio, and action prediction from one set of weights. The research team argues that no single modality gives a complete description of the world, and training on all of them at once means the modalities constrain each other. FLUX 3 builds on Self-Flow, BFL's method for aligning multimodal generation and understanding, which combines the flow matching objective with a self-supervised feature reconstruction objective. FLUX 3 Video generates clips up to 20 seconds long in a single generation with native audio, supporting text-to-video, image-to-video, video-to-video, keyframe-to-video, and generative video-audio continuation. BFL reports strong performance in human facial expressions and associating sounds with physical events. In preliminary human preference tests for 10-second text-to-video clips at 720p with audio, FLUX 3 was preferred over Luma Ray 3.2 in 93% of comparisons, over Runway Gen-4.5 in 77%, over Grok Imagine Video in 69%, over Kling v3 Pro in 60%, and over Happy Horse v1 and 1.1 in 59% and 57% respectively. Against Seedance 2.0 and Gemini Omni Flash it reached 52%. Video prediction consumes over 95% of training compute, audio under 0.5% of tokens. The same backbone runs FLUX-mimic, a robot policy under 80 ms on one RTX 5090. Access is gated: Video and Action are in early access, Image follows, open weights come last.