# Black Forest Labs launches FLUX 3, a multimodal image and video model **FLUX 3 turns a single prompt into images or 20-second videos with audio, but access is gated and pricing is still unannounced.** Black Forest Labs (BFL), the Freiburg, Germany-based lab behind the FLUX image family, expanded its lineup today with FLUX 3. The lab says the new system understands and produces images, plus combined audio-video output lasting as long as 20 seconds from one instruction, and extends the same core approach toward robotic vision and action. [VentureBeat](https://venturebeat.com/technology/black-forest-labs-launches-flux-3-capable-of-generating-images-and-20-second-video-with-audio-but-in-limited-release-to-start) reports this is BFL's first public video generation model. ## What Black Forest Labs just shipped FLUX 3 arrives as four product lines: FLUX 3 Video, FLUX 3 Image, FLUX 3 Action, and an upcoming open source FLUX 3 Dev. Only two of those, FLUX 3 Video (with optional native audio) and FLUX 3 Action, are live today, and both sit behind a gated Early Access program that anyone can apply to but that BFL itself must approve. There is no public application programming interface yet; BFL says FLUX 3 Image will arrive in the coming weeks, with general availability to follow. BFL frames the release around a single idea: rather than stitching together separate image, video, and audio systems behind one interface, FLUX 3 is jointly trained across those modalities at once. Company co-founder and CEO Robin Rombach summed up the philosophy simply: "You can't cheat reality." He argues that a model trained only on still images can only output still images, while the physical world moves, sounds, and responds. ## How FLUX 3 stacks up against video rivals BFL published early head-to-head preference results from 10-second, 720p text-to-video clips with audio, though it labels the numbers a "preliminary evaluation of an early FLUX 3 candidate" — meaning they reflect a pre-release checkpoint, not the exact model now shipping into early access. Reviewers picked FLUX 3 Video 93% of the time against Luma Ray 3.2, 77% against Runway Gen-4.5, 69% against Grok Imagine Video, and 60% against Kling v3 Pro; the two Happy Horse versions trailed FLUX 3 at 59% and 57%. Two remaining matchups matter more than the rest. Seedance 2.0's international rollout is indefinitely on hold after entertainment studios sent copyright threats, so a tie against a model most Western enterprises can't currently buy says little about live competition. Gemini Omni Flash, on the other hand, is generally available now through Google's Gemini API at $0.10 per second of generated 720p video, making it the closest apples-to-apples rival BFL has only managed to draw against, at 52%. ## What's still missing BFL has not disclosed pricing, service-level guarantees, evaluation sample sizes, rater counts, or any benchmark numbers at all for FLUX 3 Image. That leaves enterprise buyers unable to calculate total cost of ownership or independently reproduce the video comparisons before applying for access. Downloadable weights are also absent at launch. BFL says an open-weight FLUX 3 Dev model will follow later this year, described in its technical materials as covering video, audio, image, and action prediction — a wider scope than any earlier open FLUX Dev release, which covered images only. Developers used to getting a locally deployable FLUX variant alongside a major announcement will have to wait for that piece. ## Key facts - FLUX 3 is Black Forest Labs' first public video generation model, adding audio and 20-second video to its existing image capability. - FLUX 3 Video and FLUX 3 Action are in gated Early Access now; FLUX 3 Image is promised in the coming weeks. - Preference testing put FLUX 3 Video ahead of Luma Ray 3.2 (93%) and Runway Gen-4.5 (77%), and roughly even with Gemini Omni Flash (52%). - No pricing, API access, or downloadable weights exist yet; an open-weight FLUX 3 Dev model is promised later this year. - The system builds on Self-Flow, BFL's method for joint multimodal training, first publicized in March 2026. ## Check your understanding What percentage of comparisons did FLUX 3 Video win against Luma Ray 3.2? Why does a tie against Seedance 2.0 tell buyers less than a tie against Gemini Omni Flash? Which two FLUX 3 product lines are missing an announced release date or price? ## Sources - [VentureBeat: Black Forest Labs launches FLUX 3](https://venturebeat.com/technology/black-forest-labs-launches-flux-3-capable-of-generating-images-and-20-second-video-with-audio-but-in-limited-release-to-start) --- - Category: know (announcement) - Affects: video creators · enterprise developers - Published: 2026-07-23T17:58:07.000Z - Original: https://venturebeat.com/technology/black-forest-labs-launches-flux-3-capable-of-generating-images-and-20-second-video-with-audio-but-in-limited-release-to-start - Generated by tinker news