Black Forest Labs unveiled FLUX 3, a multimodal foundation model that trains jointly on images, video, and audio in a single unified architecture — the lab's bid at what it calls "real-world visual intelligence: models that perceive, predict, and act across physical and digital environments." The headline for filmmakers is FLUX 3 Video: text-to-video, image-to-video, and video-to-video generation up to 20 seconds at 720p with native, synchronized audio — including multilingual dialogue, keyframe-to-video transitions, and agentic chaining to stitch longer sequences. On the image side it synthesizes and edits across styles, aspect ratios, and resolutions with sharper complex-prompt handling and multilingual text rendering, plus fast variants aimed at cheap, rapid iteration. Underneath sits an open-weight multimodal backbone (FLUX 3 Dev) meant to serve content creation across video, audio, and image as well as action prediction. The lab even ships FLUX-mimic, a video-action model built with mimic robotics for robotic manipulation. Built on an approach BFL calls Self-Flow for aligning generation and understanding, early evaluations put FLUX 3 Video ahead of rival models in up to 77% of head-to-head comparisons, with particular strength in facial expressions, sound-to-physics association, and character consistency. FLUX 3 Video launches in early access, with the open-weight image model and backbone following afterward and action prediction offered through select partners. Update: FLUX 3 Video went generally available on August 4, reaching the API with a Draft mode for cheap iteration — and a week later the head-to-head claim got independent backing. On August 11, Arena placed an updated build at #2 on the Text-to-Video leaderboard with 1,496 points, 16 behind Gemini Omni Flash at 1,512. A first video model from an image lab landing within statistical noise of Google on blind human preference is a real result, and it arrived the same day LTX-2.5 shipped open weights. Two caveats keep it from being a coronation: the ranked build is an update BFL calls "coming soon" rather than the one you can currently call, and the open-weight FLUX 3 Dev backbone — the part that would matter most to filmmakers running locally — still has no date.