Black Forest Labs releases FLUX 3 Video and FLUX-mimic, unifying multimodal generation and robotics control

Black Forest Labs (BFL) announced FLUX 3 Video, a unified multimodal model spanning image, video, audio, and action prediction, available in early access. The model is jointly trained in a single architecture and supports capabilities such as text-to-video, image-to-video, video-to-video, generative video-audio continuation, keyframe-to-video, multilingual dialogue, and agentic chaining of clips into longer sequences.

BFL also introduced FLUX-mimic, a video-action model that combines the FLUX 3 backbone with robot learning expertise for dexterous manipulation. FLUX-mimic is designed to run on a single on-premises GPU and is already being tested with Audi in real factory settings.

The FLUX 3 architecture is explicitly framed as extendable to robotics, with BFL researchers connecting it to their earlier Self Flow research. The team claims that better video world modeling transfers directly into robot control quality and sample efficiency.

BFL’s announcement positions FLUX 3 Video as outperforming other frontier models like Seedance 2.0, Gemini Omni, and Grok Imagine in unspecified multimodal benchmarks, though these claims are not independently verified in the source.