Flux 3 adds native audio — can it finally beat Kling in the video wars?
black forest labs released flux 3, a multimodal model that generates video with native audio for up to 20 seconds, outperforming luma and runway in internal tests. this accelerates the race toward unified world models that handle robotics and cinematic generation simultaneously.
black forest labs released flux 3, a multimodal foundation model that trains on images, video, and audio simultaneously to generate 20-second clips with native sound. internal evaluations using 10-second 720p clips indicate the model significantly outperforms rivals like luma ray 3.2 (winning 93% of comparisons) and runway gen-4.5 (winning 77%), while holding a narrower edge over kling v3 pro (60%). beyond content generation, bfl introduced flux-mimic, an action model extension currently in production tests at audi, to predict robotic movements based on video understanding. this architecture leverages dedicated encoders and decoders to create a shared representation of reality, positioning bfl to compete with 'world model' efforts from deepmind and other top labs by unifying perception, prediction, and action in a single system.