Black Forest Labs launches FLUX 3, a multimodal model spanning video, audio and robotics

Black Forest Labs launched FLUX 3 on July 23, a multimodal foundation model that generates video clips with native audio up to 20 seconds long from text, images or existing video, the company said.

The release moves the German startup, known for its FLUX image generators, into video and robotics inside a single architecture jointly trained on images, video and audio. It is a direct bid for the same generative-video market that Runway, Luma, Kuaishou’s Kling and Google are racing to lead.

FLUX 3 Video is in early access now, with its robotics action-prediction capability offered initially through select partners. FLUX 3 Image is slated to arrive within weeks, and an open-weight version called FLUX 3 Dev is planned to follow. The video model supports video continuation, keyframe transitions and multilingual dialogue.

Black Forest Labs’ quality claims rest on its own tests. In 10-second, 720p preference comparisons run by the company, reviewers preferred FLUX 3 output 93% of the time over Luma Ray 3.2 and 77% over Runway Gen-4.5. The margin narrowed to 52% against both Seedance 2.0 and Google’s Gemini Omni Flash — a near coin toss, and figures the company has not had independently verified.

The robotics angle is the more unusual claim. Black Forest Labs worked with Mimic Robotics on FLUX-mimic, a video-action model tested on production tasks at Audi. It handled flexible door-seal assembly with a response time of roughly 101 milliseconds per action, according to Audi — fast enough, in principle, to drive a robot arm on a live line rather than in a lab demo.

Whether a single model can serve both filmmakers and factory robots is the open question. The open-weight FLUX 3 Dev release, still to come, will be the clearer test of whether developers adopt the architecture beyond Black Forest Labs’ own benchmarks.