Black Forest Labs unveils FLUX 3 AI: ditching still video footage and robotic hands

Black Forest Laboratories was launched Flow 3 On Thursday, for the first time, the company’s flagship model creates video instead of just still images. The German AI lab, known for its FLUX line of image generators, trained the new system on images, video and audio simultaneously, within a single shared system.

This is what’s known as multimedia: a single model that learns several types of information together rather than through separate tools installed side by side.

The video aspect is the main feature. FLUX 3 produces clips up to 20 seconds long, with the sound created alongside the image and synchronized with what’s happening on screen — dialogue, sound effects, and ambient noise. In early evaluations, human reviewers preferred the FLUX 3’s output over the Runway Gen-4.5 in 77% of head-to-head comparisons and over the Luma Ray 3.2 in 93%. It appears to be slightly better than the Gemini Omni and Seedance, beating those models in 52% of evaluations.

Of course, this is a preference test, not a fixed evaluation criterion: evaluators simply watch two clips and choose the one that looks more convincing, and the BFL counts how many times FLUX 3 wins.

Other than that, the model seems to be very competent for stills as well, following its legacy. BFL shared some images, and it looks like the FLUX 3 is versatile and capable of generating a wide range of styles that go beyond photorealism.

BFL frames this as more than just a content tool. “A model that only learns images can only generate images,” said co-founder and CEO Robin Rumbach. The company’s bet is that learning to predict video also means learning the physics behind it — weight, contact, timing — which is exactly what a machine needs to move through the physical world.

This bet has a name: Flow imitation. Built using Zurich-based Simulated Robotics, it takes FLUX 3’s video prediction engine and adds a lightweight “decoder” — a small plug-in that translates the model’s internal sense of how things move into actual robot movements. Automaker Audi is already testing this system on tasks such as installing flexible door seals, work that traditional automation struggles to handle.

“Audi represents the kind of manufacturing partner for whom we built FLUX-mimic,” said Stefan Daniel Gravert, co-founder of mimic. Audi’s Christoph Schneider said robots now “solve complex manipulations of soft objects” that older machines cannot touch. BFL says the complete system reacts in about 101 milliseconds, in the neighborhood of human visual feedback.

FLUX’s rise did not happen in a vacuum. Founded in August 2024 by veteran researchers who helped build the original Stable Diffusion models in Stability AI, Black Forest Labs launched Flux models that outperformed MidJourney and outperformed the disappointing Stability. Stable spread 3.

The open source models Flux Dev and Schnell have earned the title of “Best Open Source Image Generator,” which AI artists expect Stable Diffusion 3.5 will eventually reclaim.

It never happened. FLUX 1.1 Pro went to the top Synthetic image analysis arena In October of that year. This wasn’t open source, though.

BFL issued FLUX.2 inches November 2025 But it was not popular. The open source crown held by the original Flux has continued all the way to Alibaba Z-Image Turbo It was phased out in late 2025, to match its quality on lower-end consumer graphics cards. “This is what SD3 was supposed to be,” one CivitAI user wrote at the time.

FLUX 3 is the return of BFL, and it’s not fully opened yet. Video and Motion are now in early access through APIs and select partners, and are simulating bots among themselves, with image creation continuing “in the coming weeks,” according to BFL. The open-weight Dev Edition, which is the only tier BFL plans to release for local use, is not scheduled for release until later in 2026.