Member of Technical Staff

Black Forest Labs opens FLUX 3 Video API for 20-second Full HD clips. Key points * FLUX 3 Video is available through Black Forest Labs' dashboard and API, generating clips up to 20 seconds in HD or Full HD.[[1]] * Text-to-video and image-to-video cost $0.17 per second in HD or $0.29 per second in Full HD; synchronized audio is included.[[2]] * BFL's ranking above Seedance 2.0 and Gemini Omni Flash is a vendor benchmark, and its earlier head-to-head tests showed only a 52% preference for FLUX 3 against each rival.[[3]] Black Forest Labs has made FLUX 3 Video available through its dashboard, API and partner platforms, moving its first video-generation model beyond the early-access program announced on July 23. The system produces HD or Full HD clips lasting up to 20 seconds and can generate dialogue, sound effects and ambience alongside the frames.[[1]] The release supports text-to-video, image-to-video, keyframe transitions and video continuation. BFL also promotes multi-scene generation, typography embedded in moving scenes and multilingual dialogue with synchronized lip movement. Its public product and API pages, however, describe the speech support as multilingual without publishing a tested language list, so the reported claim of more than 14 languages remains vendor-supplied.[[1]] [[4]] Longer clips, familiar competitive claims. FLUX 3 costs $0.17 per second for standard HD text-to-video or image-to-video and $0.29 per second for Full HD. That puts a 20-second generation at $3.40 or $5.80, respectively. Draft-mode HD costs $0.06 per second, and audio carries no additional charge.[[2]] BFL's Elo tables rank FLUX 3 first for text-to-video and image-to-video, ahead of Seedance 2.0 and Google's Gemini Omni Flash.[[4]] Those results are not independent benchmarks. In BFL's preliminary early-access comparison of 10-second, 720p clips, raters preferred FLUX 3 over Seedance and Gemini Omni in 52% of comparisons, close to an even split.[[3]] The feature list also overlaps with those rivals. ByteDance says Seedance 2.0 generates multi-shot audio-video clips up to 15 seconds and accepts text, images, video and audio references.[[5]] Google's preview Gemini Omni Flash creates three- to 10-second 720p videos with audio, readable text and conversational editing.[[6]] FLUX 3's clearest documented distinction is its longer single-generation limit, rather than an independently established quality lead. Companies mentioned. At the intersection of AI, tech, and markets. The stories that matter, in one email. Free - unsubscribe anytime.

Black Forest Labs releases FLUX 3: A multimodal flow model for image, video, audio and robot action prediction. July 26, 2026 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 also the first FLUX model to ship video, audio and action prediction from one set of weights. The Black Forest Labs (BFL) research team argues that no single modality gives a complete description of the world. Images capture spatial structure at one instant. Video restores time and exposes physical dynamics. Audio reveals causal relationships between mechanical events and sound. Each is treated as a lossy projection of the same underlying reality. Training on all of them at once means the modalities constrain each other. The sound has to match the impact. The motion has to obey the mass. The research team calls FLUX 3 its first model built entirely on that principle. The method underneath: Self-Flow. FLUX 3 builds on Self-Flow, BFL's method for aligning multimodal generation and understanding in one architecture. Self-Flow combines the flow matching objective with a self-supervised feature reconstruction objective. The reference implementation on GitHub is Apache-2.0 and uses SiT-XL/2 with per-token timestep conditioning. It trains with a 25% per-token mask ratio and self-distillation from an EMA teacher at layer 20 to a student at layer 8. That released checkpoint is an ImageNet 256x256 research model, not FLUX 3. BFL states that it 'significantly scaled up compute and data resources' on the same approach to train FLUX 3 across video, images and audio simultaneously. Self-Flow itself was introduced in March 2026, so it is not new to this launch. What is new is the scale. What FLUX 3 Video does. FLUX 3 Video generates clips up to 20 seconds long in a single generation, with native audio. The supported modes cover text-to-video, image-to-video, video-to-video from a reference clip, keyframe-to-video for controlled transitions, and generative video-audio continuation from input video and audio. BFL also lists multilingual dialogue, agentic chaining of clips into multi-shot sequences, and strong typography generation with animated designs. The BFL team reports particular strength in human facial expressions and in associating sounds with physical events. Performance. BFL team published preliminary human preference results. The setup was 10-second text-to-video clips at 720p with audio. FLUX 3 was preferred over Luma Ray 3.2 in 93% of comparisons and over Runway Gen-4.5 in 77%. Against Grok Imagine Video the figure is up to 69%, then Kling v3 Pro at 60%, Happy Horse v1 at 59% and Happy Horse 1.1 at 57%. Against Seedance 2.0 and Gemini Omni Flash the result is 52%, close to a coin flip. Interactive explorer. bfl@ flux-3 :~/real-world-models Early Access flux3 $ flux3 -spec One backbone, four output types. * Model class - Multimodal flow matching foundation model * Joint training modalities - image + video + audio, one architecture * Extended output - Action prediction (robot state) * Max video length - 20 seconds in a single generation * Audio - Native, generated jointly with video * Video modes - text-to-video, image-to-video, video-to-video, keyframe-to-video, video+audio continuation * Other capabilities - multilingual dialogue, agentic multi-shot chaining, typography and animated design * Announced - 23 July 2026, Black Forest Labs * Pricing - not published BFL describes each modality as a lossy projection of one underlying reality. Training on all of them at once lets their mutual constraints supervise each other. flux3 $ enter

FLUX 3: one AI model for Video, audio, and factory robots. On July 23, 2026, Black Forest Labs (BFL) launched FLUX 3, and the model is unlike anything the company has released before. Where earlier FLUX models stopped at still images, FLUX 3 is trained simultaneously on images, video, and audio inside a single architecture. The same backbone that generates a 20-second clip with synchronized audio is also driving factory robots at Audi. For enterprise teams building creative workflows or physical AI systems, that convergence is the story worth understanding. One architecture, four product lines. BFL frames FLUX 3 around a thesis it calls "visual intelligence": models that can perceive, predict, and act across physical and digital environments. The claim is that a model must learn a representation of the world, not just its still-frame snapshots, to generate convincing video or reliable robot actions. FLUX 3 is the first public result of that research direction. The model ships in four product lines: | Product | Capability | Status (July 25, 2026) | | FLUX 3 Video | 20-second clips with synchronized audio, text-to-video and image-to-video | Gated early access | | FLUX 3 Image | Advanced image synthesis and editing | Coming in weeks | | FLUX 3 Action | Action prediction for robotics, starting with FLUX-mimic | Gated early access | | FLUX 3 Dev | Open-weight multimodal backbone (video, audio, image, action) | Later in 2026 | Early access for Video and Action is available by application at bfl.ai. FLUX 3 Image and open-weight Dev access follow on rolling timelines that BFL has not pinned to specific dates. What FLUX 3 Video actually does. FLUX 3 Video generates clips up to 20 seconds long, with audio produced alongside the visual content and matched to what happens on screen: dialogue, sound effects, ambient noise. In early human-preference evaluations, FLUX 3 Video was preferred over Runway Gen-4.5 in 77% of head-to-head comparisons and over Luma Ray 3.2 in 93%, according to BFL's launch materials. The model also handles multilingual dialogue and supports agentic chaining of individual clips into longer multi-shot sequences while maintaining character and visual reference consistency. Enterprise creative teams running campaigns at scale should note the workflow implication: a single generation call can cover the video, its audio track, and the keyframe images from which both can be edited, without handing off between separate vendor APIs. The company's existing distribution makes that consolidation credible. Earlier FLUX models power generative features in Adobe Photoshop, Picsart, and Nous Research's Hermes Agent. More than 500 million downloads of FLUX models have been recorded to date, per the company's press release. FLUX 3 Video is already in early testing with Canva, Burda, Magnific (formerly Freepik), Krea, and Picsart. Physical AI: from content to factory floor. The part of FLUX 3 that signals a longer strategic bet is FLUX-mimic. BFL partnered with Swiss robotics firm mimic robotics to build a video-action model on top of the FLUX 3 backbone. The model is being tested and deployed in Audi production facilities for tasks that have resisted conventional automation: fitting flexible door seals, kitting parts into structured trays, inserting electronic control units into tight-fitting fixtures, and handling soft, flexible materials. The efficiency claim is the headline: FLUX-mimic can be fine-tuned for a specific manipulation task with as little as 30 minutes of robot data. Prior approaches required 30 or more hours. BFL attributes this to the FLUX 3 backbone already encoding how the physical world behaves, so the robot only needs to learn how a specific task maps onto knowledge it already holds. The full system reacts in approximately 101 milliseconds, in the range of human visual reaction time. "We have seen these robots solve complex soft-body manipulation work that would have been simply impossible with conventional robotics," Christoph Schneider of Audi Production Lab said in BFL's announcement. "This can have a major impact in assisting our employees, increasing efficiency, and expanding flexible automation across production and logistics operations." For enterprise buyers outside manufacturing, the Audi deployment is a proof point worth tracking: it is the first public case of a foundation model trained on consumer video content being transferred, with minimal task-specific data, into a production industrial setting. Why the unified architecture matters for enterprise buyers. Most enterprise creative stacks today involve four to six distinct vendors: one for image generation, one for video, one for audio, one for 3D rendering or product visualization, and so on. Each vendor requires its own integration, contract, and fine-tuning dataset. FLUX 3's unified architecture is a direct challenge to that structure. VentureBeat's coverage quotes BFL's pitch to enterprise software companies: "A single foundation could potentially support storyboarding, image editing, product rendering, video variation and localization without repeatedly translating assets and instructions between disconnected models." For GTM teams and agencies running content at scale, the practical benefit is fewer round-trips and less context loss when an asset moves from concept to finished format. For AI-native platforms embedding generative features, a single fine-tune on FLUX 3 that updates image, video, and audio quality simultaneously is meaningfully cheaper than maintaining separate model versions. This connects to a pattern Enera covered in the broader enterprise AI price war: as foundation model providers consolidate capabilities into fewer endpoints, the cost-per-workflow calculation shifts in ways that favor unified vendors over specialist ones. What is still missing. FLUX 3 launched without several things enterprise buyers typically need before committing: * No public pricing. BFL has not announced API costs, enterprise tiers, or SLAs. * No full benchmark methodology. Published benchmark results are expected alongside broader availability. * FLUX 3 Dev details are incomplete. License terms, parameter counts, quantization options, and hardware requirements for the open-weight version have not been disclosed. * FLUX 3 Image is not live yet. This is likely the most immediately useful tier for the creative teams already using FLUX 1 and FLUX 2. None of these gaps disqualify FLUX 3 as a story worth following. But they do mean that enterprise procurement conversations are premature until BFL releases pricing and the FLUX 3 Image tier goes live. What enterprise teams should watch. If you are running creative operations, product marketing, or physical AI initiatives, three things are worth tracking in the near term: * FLUX 3 Image launch. This is the earliest point at which most enterprise creative teams can evaluate the model in practice, not just in press materials. * FLUX 3 Dev license terms. The license will determine whether enterprises can run the model locally for sensitive workflows, which is how earlier FLUX Dev releases drove wide adoption. As Enera noted in its look at enterprise AI token efficiency, local deployment is increasingly a compliance and cost requirement, not just a preference. * Independent benchmark reproduction. BFL's own evaluations show strong results. Independent testing of FLUX-mimic's sample efficiency claims and FLUX 3 Video's head-to-head numbers will determine whether those numbers hold in realistic deployment conditions. BFL is valued at $3.25 billion and has raised more than $450 million from investors including a16z, NVIDIA, Salesforce Ventures, Adobe Ventures, Figma Ventures, Canva, and Deutsche Telekom's T.Capital. The investor list mirrors the distribution network: the platforms that invest in BFL also embed FLUX models into their products. That alignment makes FLUX 3 more likely to reach enterprise creative stacks quickly than a comparable model from a lab without those ties. If you want to evaluate how a unified visual AI foundation fits your current creative or physical AI stack, let's talk.