FLUX 3: Black Forest Labs' Unified Image, Video, and Robotics Model
What is FLUX 3?
FLUX 3 is a single multimodal frontier model from Black Forest Labs that generates images, creates up to 20-second video-with-audio clips, and predicts robotic actions — all from one prompt and one underlying architecture. Unveiled July 23–25, it collapses a boundary that has held since AI models first split into separate image, video, and robotics categories.
What FLUX 3 can do right now
FLUX 3's core capability is joint training across image, video, and audio, then an extension that repurposes the same architecture for physical action prediction. That second capability — marketed under the name FLUX-mimic — is already deployed on Audi production lines, using visual-action prediction to guide robotic tasks on the factory floor.
This isn't a research demo: a model built for creative image and video generation is performing physical work in an automotive plant days after its public unveiling.
Availability: what's live, what's coming
- FLUX 3 Video and FLUX 3 Action are live in gated early access as of July 25
- FLUX 3 Image is expected in the coming weeks
- An open-weight "Dev" backbone is planned for later in 2026, following BFL's usual pattern of releasing a research-friendly variant after the flagship launch
Why FLUX 3 matters for the future of visual AI
Most "multimodal" models today are really separate models stitched together behind one API. FLUX 3's differentiator is a single trained network spanning image, video, audio, and physical action — a shift that, if it holds up beyond the Audi pilot, positions Black Forest Labs less as an image-generation vendor and more as a foundation-layer player in what it calls "visual intelligence."
That places FLUX 3 in a different competitive lane than the video-editing tools currently crowding the market. As FLUX 3 Image rolls out more broadly and the open-weight Dev variant arrives later in 2026, it will be worth watching whether this unified architecture proves scalable across the broader robotics and generative media landscape.