Skip to main content
Back

Black Forest Labs' FLUX 3 Generates Video, Audio, and Drives Audi Robots

Black Forest Labs launched FLUX 3, one model that jointly generates images, 20-second video with synchronized audio, and robot actions from a shared backbone.

  • FLUX 3 from Black Forest Labs jointly generates image, video, audio, and robot actions from one backbone.
  • Video mode produces 20-second clips at 720p with native synchronized audio in a single generation.
  • Preferred over Luma Ray 3.2 in 93% of tests and Runway Gen-4.5 in 77%, per preliminary results.
  • FLUX-mimic robotics variant runs live on Audi production lines for soft-body manipulation tasks.
  • Fine-tunes on 30 minutes of robot data versus 30+ hours for prior approaches; 101ms latency.
  • Video and Action in gated early access; Image ships soon; open-weight FLUX 3 Dev arrives later in 2026.

Black Forest Labs, the German lab whose FLUX.1 image models became a default in open-weight generation, has moved well beyond still pictures. FLUX 3 is a single multimodal model that jointly generates images, up to 20 seconds of video with synchronized audio, and predicts robot actions, all from one shared architecture rather than a stitched-together stack of specialized systems.

The Freiburg-based lab says FLUX 3 is jointly trained across those modalities rather than assembling separate image, video, and audio models behind a common interface. That framing is the entire pitch. The same weights learn every modality at once, then extend to predict physical actions for robots, instead of bolting audio onto a video model or wrapping an image generator in a video head.

Four products, staggered release

FLUX 3 ships as four separate product lines with very different availability. Video and Action are gated early access today, Image follows shortly, and the open-weight version arrives last.

  • FLUX 3 Video: text-to-video, image-to-video, video-to-video, keyframe transitions, and generative continuation, all with native audio
  • FLUX 3 Action: robot action prediction via early access
  • FLUX 3 Image: expected in the coming weeks
  • FLUX 3 Dev: the open-weight multimodal backbone, planned for later in 2026

No pricing has been announced for any tier, a real gap for anyone trying to plan a budget. There is also no published SLA, no image benchmarks, and no rater counts for the preference tests the company did share.

What the video model actually does

FLUX 3 Video generates clips up to 20 seconds long in a single generation, with native audio. Supported modes cover text-to-video, image-to-video, video-to-video from a reference clip, keyframe-to-video for controlled transitions, and generative video-audio continuation from input video and audio. BFL also lists multilingual dialogue, agentic chaining of clips into multi-shot sequences, and strong typography generation with animated designs.