Crypto
Home›Crypto›Market Structure›Black Forest Labs launches FLUX 3, its first video-gen…
Black Forest Labs launches FLUX 3, its first video-generating AI model
The early-access release lets FLUX 3 create clips up to 20 seconds long with synced audio, and the same system is being used to train robot work on an Audi production line.
Black Forest Labs has launched FLUX 3 in early access, its first model that generates video rather than only still images. The system can produce clips up to 20 seconds long, with audio generated alongside the visuals and synced to on-screen action, including dialogue, sound effects, and ambient noise.
The company positioned FLUX 3 as a multimodal system trained on images, video, and audio within a shared model. In early evaluations, human reviewers preferred FLUX 3 to Runway Gen-4.5 in 77% of head to head comparisons, and to Luma Ray 3.2 in 93%, while also beating Gemini Omni and Seedance in 52% of evaluations.
Black Forest Labs said it is already applying the same backbone to robotics through FLUX-mimic, a model built with mimic robotics that uses FLUX 3’s video prediction engine plus a lightweight decoder to translate motion understanding into robot actions. Decrypt reports the co-founder and CEO Robin Rombach framed the approach as teaching more than content generation, arguing that video prediction can help a model learn underlying physics such as weight, contact, and timing.
On availability, only the open-weight “Dev” version of FLUX 3 is planned for later in 2026, while Video and Action remain behind APIs and partner access for now, with Image expected in the coming weeks. Audi is already testing FLUX-mimic on tasks such as fitting flexible door seals.