Black Forest Labs, the German lab behind the FLUX image models, has gone frontier. On 23 July it unveiled FLUX 3, its first attempt at one model that handles image, video, audio and even robot actions, all trained together rather than bolted side by side.
Quick jargon check. Multimodal here means a single model that works across several media (pictures, moving images, sound) instead of separate tools stitched behind one button. Open weights means the trained model file itself is released so you can download and run it on your own machine. Black Forest Labs made its name on the open-weight FLUX image models that power a chunk of the AI art you have seen, so this matters to a lot of people who build on top of it.
What FLUX 3 claims to do
The headline feature is FLUX 3 Video: up to 20 seconds of footage with native audio, generated from text, a still image, keyframes or a reference clip. It can carry a character across scenes, do multilingual dialogue, and chain shots into longer sequences. There is also a robotics component, FLUX-mimic, that predicts robot actions, which is Black Forest Labs planting a flag in physical AI.
Read the rollout, not the press release
Here is where you keep a hand on your wallet. FLUX 3 Video is in gated early access, meaning you apply at bfl.ai and wait for approval, with access running through the API and private weights for selected partners. FLUX 3 Image lands in the coming weeks. And FLUX 3 Dev, the open-weight release that made the FLUX name in the first place, is planned for later in 2026.
So the thing the open-source crowd actually wants, the downloadable model, is last in the queue, while the shiny gated video demo goes first. That is a very 2026 move: lead with the buzz, meter the access, and let the community goodwill built on the free stuff sell the paid stuff. It might all arrive as promised. It might not, and until independent testers can run it, that 20-second audio-and-video claim is a claim, not a fact.
Still, the direction is real and it is a threat to the incumbents. If a lab known for open weights can land credible video with sound, the small studios and solo creators who cannot stomach top-tier subscription prices finally get a serious option they might one day self-host. That is the part worth watching, and it is the part Black Forest Labs is making you wait longest for.
If you make video for a living, get your application in now so you are not stuck at the back of the queue. If you were here for the open weights, temper the excitement until later in 2026 turns into an actual download link.
Did you know: Black Forest Labs was founded by researchers who helped create Stable Diffusion, the model that kicked off the open-source image-generation boom in the first place.