Key facts
- Black Forest Labs released FLUX 3, its first AI model capable of generating video.
- FLUX 3 produces video clips up to 20 seconds long with synchronized audio.
- Human reviewers preferred FLUX 3's video output over competitors like Runway Gen-4.5 and Luma Ray 3.2.
- The technology behind FLUX 3 is also used in FLUX-mimic for robotics applications.
- Audi is testing FLUX-mimic on its production line for tasks like fitting door seals.
- The open-weight 'Dev' version of FLUX 3 is scheduled for release in late 2026.
Black Forest Labs has launched FLUX 3, marking its entry into video generation with clips up to 20 seconds long and synchronized audio. The German AI lab, previously known for its image generation models, has developed FLUX 3 as a multimodal system trained on images, video, and audio simultaneously.
In early evaluations, FLUX 3's video output was favored over competitors such as Runway Gen-4.5 and Luma Ray 3.2. The model also demonstrates strong capabilities in generating still images across various styles. The company views this advancement as more than just a content creation tool, believing that learning to predict video also imparts an understanding of physics relevant to physical world interactions.
This principle underpins FLUX-mimic, a robotics model built in collaboration with mimic robotics. FLUX-mimic utilizes FLUX 3's video prediction engine, enhanced with a decoder to translate motion predictions into robot actions. Automaker Audi is currently testing this system on its production lines for tasks like fitting flexible door seals, work that has proven challenging for traditional automation. Audi has noted that the robots can now handle complex soft-body manipulation, with the system reacting in approximately 101 milliseconds.
Black Forest Labs, founded in August 2024 by former Stability AI researchers, has a history of developing competitive AI models. Their earlier open-source image generators, Flux Dev and Schnell, were noted for their quality. FLUX 3's release is positioned as a comeback, though its advanced video and action generation features are currently behind APIs and partner access. The company plans to release an open-weight 'Dev' version in late 2026.