Unified multimodal model for images, video with synchronized audio, and robotic action. FLUX.3 Video makes 20-second clips; Image does synthesis and editing; Action predicts robotic manipulation. All use flow matching.
Released Jul 2026. Video/action in early access; image and open-weight to follow.
Updated 23 Jul 2026Checked 10 Oct0 updates this week
Learning from video and audio builds models of physical behavior. Supports images, video with synced audio, and robotic action in one system.
Replaces diffusion with learnable paths, reducing computation and inference time across all modalities.
Keyframe-controlled transitions, image-to-video with character consistency, video continuation, and dialogue control.
| Spec | FLUX.3 |
|---|---|
| Maximum video duration | 20 secondsR (reported) |
| Native audio support | Multilingual dialogue, effects, ambienceR (reported) |
| Multi-year partnership contracts | $400M+R (reported) |
R reported by the company
Black Forest Labs
FLUX, image generation, video generation, flow matching
German AI company by latent diffusion researchers. FLUX multimodal models for images, video, audio and robotics using flow matching. Partnerships exceed $400M.
