Black Forest Labs opens FLUX 3 early access, with its bigger robotics case still to prove
Black Forest Labs has opened gated early access to FLUX 3, a model it says generates image and video with audio and can underpin robot-action work. Its published video results are preliminary company preference tests, while the claimed robotics data advantage has not been independently detailed.
- FLUX 3 is a gated early-access release, not yet a generally available video service or an open-weight model.
- Black Forest Labs says it can make video with native audio for up to 20 seconds and reuse its video backbone for robot-action models.
- The public evidence so far is a preliminary internal video preference test and a partner-reported robot-data claim.
Black Forest Labs has opened early access to FLUX 3, its first public video-generation model. The Freiburg, Germany-based lab was founded by researchers behind latent diffusion, Stable Diffusion and the earlier FLUX image models; co-founder and chief executive Robin Rombach, who helped create Stable Diffusion, is presenting the release as a step beyond image generation, toward models that can handle image, video and audio together.
The immediate capability is concrete but controlled. Black Forest Labs says FLUX 3 can make a video with native audio for up to 20 seconds from text or supplied image and video references. The product page asks prospective users to request early access, and a launch report says there is no public API access yet.
The larger claim is harder to establish: that a common model of visual dynamics can serve both creative tools and factory robots. That is an architectural proposition with commercial consequences, not a demonstrated equivalence between making a convincing clip and running a production line.
One backbone, four planned product lines
In its launch announcement, Black Forest Labs calls FLUX 3 its first model trained jointly on images, video and audio. The company says the modalities impose useful constraints on one another: motion should reflect an object's apparent mass and sound should correspond to an event.
That shared representation is the distinction Black Forest Labs is asking customers to value. FLUX 3 Video and FLUX 3 Action have entered early access; Image is due after its own early-access phase; and FLUX 3 Dev is planned as an open-weight multimodal backbone for content creation and action prediction. The company says the eventual Video and Image products will be offered through APIs and private-weight access.
For now, the sequence matters more than the roadmap. FLUX 3 has not launched with downloadable weights or a license, and neither its announced price nor a production service-level commitment has been published. Those absences leave customers unable to compare its cost, deployment control and reliability against video systems they can already buy.
