Black Forest Labs

Generative AI · Video Generation · Robotics

Launch

Black Forest Labs launches FLUX 3, unifying video, audio, robot control

By training all modalities together, the model claims to have learned physical cause-and-effect well enough to already run robots on a live car factory line.

• Black Forest Labs opened early access to FLUX 3 on July 23, 2026, a multimodal foundation model that generates images and up to 20-second video clips with synchronized native audio from a single prompt.
• Freiburg, Germany-based Black Forest Labs built FLUX 3 as its first public video generation model, extending a FLUX family whose image tech already powers Adobe Photoshop and Picsart.
• FLUX 3 rolls out across four product lines: FLUX 3 Video, FLUX 3 Image, FLUX 3 Action, and an upcoming open-weight FLUX 3 Dev for developers.
• In early benchmarks FLUX 3 was preferred over Runway's Gen-4.5 in 77% of comparisons, while video prediction alone consumed over 95% of the model's total training compute.
• A collaboration with mimic robotics produced FLUX-mimic, a video-action variant already deployed on Audi production lines to handle dexterous tasks like inserting and manipulating soft materials.
• The release caps a fast cadence: FLUX 1 shipped in August 2024, FLUX 2 in November 2025, and FLUX 3 arrives roughly eight months later as competition in video generation intensifies.
• Black Forest Labs was last valued at $3.25 billion after a $300 million Series B, giving it outsized resources and attention as it pushes into video, audio and robotics simultaneously.
• By jointly training on video, audio and robot actions instead of stitching separate models together, BFL is betting a single 'world model' can serve both content creation and physical automation, a wager that could blur the line between media AI and robotics markets.