Black Forest Labs released FLUX 3, a multimodal AI model that generates images and video with audio in clips up to 20 seconds from single text prompts. The Freiburg-based lab trained the model jointly across image, video, and audio modalities rather than combining separate specialized models, a distinction the company says enables better coherence across output types.

FLUX 3 extends beyond traditional creative generation. Black Forest Labs positions the architecture as applicable to robotic vision and control, treating image generation, video synthesis, computer use simulation, and robotics as variations of a unified "visual intelligence" capability. This architectural choice differs from competitors who typically stitch together independent models optimized for each task.

The release comes in limited availability initially. Black Forest Labs has not specified rollout timelines or pricing details for broader access. The limitation reflects both the computational cost of inference and standard practice among frontier model labs testing system behavior before scaling.

The company's framing matters. By arguing that image generation and robotic action control share underlying principles, Black Forest Labs positions FLUX 3 as a foundation for embodied AI applications rather than a creative tool alone. This interpretation aligns with industry moves toward unified architectures that handle multiple modalities and task types through a single learned representation.

Technical details remain sparse. The company has not disclosed training dataset composition, model size, or comparative benchmarks against existing image and video models. Performance on the 20-second video generation claims, which represent a significant extension from FLUX's previous image-only capabilities, will determine whether joint training actually delivers advantages over ensemble approaches.

FLUX 3 enters a crowded field. OpenAI's GPT-4o, Anthropic's Claude, and Google's Gemini all handle multiple modalities. Video generation specifically faces established competition from Runway, Pika, and others. Black Forest Labs' bet rests on whether