Black Forest Labs brings FLUX 3 AI into video and robotics

Black Forest Labs released FLUX 3 AI in early access Thursday, moving its flagship generative model from still images into video and robotics. For everyday investors tracking the AI race, the launch shows how image-generation labs are pushing into bigger markets: media tools, industrial automation and physical-world AI.

The German AI company said FLUX 3 can create videos of up to 20 seconds with audio generated alongside the visuals. That includes dialogue, effects and background sound aligned with what appears on screen, according to Black Forest Labs.

FLUX 3 was trained on images, video and audio inside one shared system, the company said. That approach is called multimodality, meaning one model learns multiple kinds of data together rather than relying on separate systems for each format.

What is FLUX 3 AI?

FLUX 3 AI is Black Forest Labs’ newest generative model and the first in its FLUX line built to generate video as well as images. The company said the model is designed as a broader foundation for visual creation and real-world motion, rather than only a tool for producing still pictures.

Black Forest Labs said human reviewers preferred FLUX 3’s video output over Runway Gen-4.5 in 77% of head-to-head comparisons and over Luma Ray 3.2 in 93% of comparisons. The company also said FLUX 3 beat Gemini Omni and Seedance in 52% of those evaluations.

Those numbers come from preference testing, not a fixed technical score. In that kind of test, reviewers compare two outputs and choose which one they find more convincing, then the company reports how often its model was selected.

The company also showed still-image examples and said FLUX 3 remains capable across a range of visual styles. That matters because Black Forest Labs built its name around image generation before adding video to the model line.

How does FLUX 3 connect to robot hands?

Black Forest Labs is also using the same model backbone for FLUX-mimic, a robotics system built with Zurich-based mimic robotics. The companies said FLUX-mimic uses FLUX 3’s video-prediction engine and adds a lightweight decoder, which translates the model’s understanding of motion into robot actions.

Black Forest Labs co-founder and CEO Robin Rombach linked the video work to robotics, saying, “A model that only learns images can only generate images.” The company’s argument is that video prediction can help a model learn timing, contact and movement, which are needed when machines interact with physical objects.

Audi is already testing FLUX-mimic on its production line, including work such as fitting flexible door seals, according to Black Forest Labs. mimic co-founder Stephan-Daniel Gravert said Audi is the kind of manufacturing partner the system was built for, while Audi’s Christoph Schneider said the robots can handle “complex soft-body manipulation work” that older automation could not.

Black Forest Labs said the full robotics system responds in about 101 milliseconds, a speed it compared to human visual reflexes.

Who can use FLUX 3 now?

Access is still limited. Black Forest Labs said only the open-weight “Dev” version is planned for later in 2026. Open-weight means developers can access the model’s weights instead of only using it through a hosted product.

For now, the Video and Action systems remain available through APIs and partner access, according to the company. The Image version is expected to follow in the coming weeks.

This story draws on original reporting from Decrypt.