FLUX 3

FLUX 3 jointly learns from images, videos, and audio within a single architecture to build a comprehensive representation of the world.

All video outputs generated by FLUX 3 come with integrated, native audio generation, associating sounds with physical events.

The model can create diverse videos up to 20 seconds in length directly from pure text prompts.

FLUX 3 can generate videos by continuing from a starting image (animation) or by using images as visual references.

FLUX 3's world understanding extends to action prediction, either natively integrated or as a backbone for finetuning specialized action models for robotics.

FLUX 3 can synthesize and edit images across a wide variety of styles, aspect ratios, and resolutions, including high-accuracy text rendering in multiple languages.

The model supports agentic chaining of individual clips into longer, multi-shot sequences, with visual references helping to ensure character consistency across scenes.

FLUX 3 handles a broad range of visual styles from candid camcorder footage to animation and cinematics, and supports multilingual dialogue in video outputs.