Home / Technology / AI Sees, Learns, Acts: New Multimodal Model Launches
AI Sees, Learns, Acts: New Multimodal Model Launches
23 Jul
Summary
- New AI model understands and generates images, audio, and video.
- Architecture extends to robotic vision and actions for physical tasks.
- Launches with early access for video and action generation features.
Black Forest Labs (BFL) has unveiled FLUX 3, a novel multimodal frontier model trained to comprehend and generate images, as well as audio/video clips up to 20 seconds from a single prompt. This unified architecture extends to robotic vision and actions, enabling AI to perceive, predict, and act across both physical and digital landscapes. FLUX 3 represents BFL's inaugural public video generation model.
The model will be available through FLUX 3 Video, FLUX 3 Image, FLUX 3 Action, and the forthcoming FLUX 3 Dev. FLUX 3 Video, with optional native audio generation, and FLUX 3 Action are now accessible via an "Early Access" program. While specific pricing and comprehensive benchmarks are yet to be released, BFL has indicated that open-weight versions are planned for later this year.
FLUX 3 builds upon BFL's Self-Flow method, aligning multimodal understanding and generation within a single architecture. The company emphasizes that training across video, images, and audio simultaneously demonstrated that video generation and action prediction do not require separate foundations. This approach targets creative tooling, media, design, e-commerce, and physical AI applications.
The video generation capabilities include text-to-video, image-to-video, and video-to-video generation, with outputs of up to 20 seconds including native audio. For robotics, FLUX-mimic, built on FLUX 3, aims to significantly reduce the data required for robots to learn new manipulation tasks, potentially cutting training time from hours to minutes.
