Published: July 24, 2026
Black Forest Labs announced FLUX 3 on July 23, 2026, a new multimodal frontier model that is jointly trained on images, video, audio and action prediction within a single unified architecture. FLUX 3 is the company’s first public video generation model, producing clips of up to 20 seconds with synchronized audio, and it also powers FLUX-mimic, a robotics model already being tested by Audi. FLUX 3 Video and FLUX 3 Action are available in early access, with FLUX 3 Image and open-weight versions to follow.
What did Black Forest Labs release today?
Black Forest Labs, the Freiburg-based AI lab founded by the researchers behind latent diffusion and Stable Diffusion, introduced FLUX 3 as the newest addition to its FLUX family of visual AI models. Unlike earlier FLUX releases, which focused on image generation and editing, FLUX 3 learns from images, video and audio at the same time within one architecture, and the same backbone can be extended to predict physical actions. The model builds on Self-Flow, the company’s approach for aligning multimodal generation and understanding in a single architecture.
The release covers four variants:
- FLUX 3 Video,
- FLUX 3 Image,
- FLUX 3 Action,
- FLUX 3 Dev.
Together with robotics company mimic robotics, Black Forest Labs also introduced FLUX-mimic, a video-action model built on FLUX 3 that is aimed at general-purpose robotic manipulation.
What can FLUX 3 do?
FLUX 3 Video is Black Forest Labs’ first public video generation model. It produces clips of up to 20 seconds, with audio generated alongside the picture and synchronized to what happens on screen, including dialogue, sound effects and ambient noise. According to Black Forest Labs, the model is particularly strong at capturing human facial expressions, associating sounds with physical events, and handling multiple languages.
Because the model is trained across modalities together, the same foundation supports image generation and editing, video with native audio, and action prediction. Black Forest Labs positions this as one underlying capability, visual intelligence, that spans generative media, simulation and physical AI. The company says testing showed that video generation and action prediction do not require separate foundations: the video-trained architecture could be extended to action prediction without losing its video capabilities.
How does FLUX 3 compare to Runway Gen-4.5 and Luma Ray 3.2?
Black Forest Labs reports that in early human evaluations, evaluators preferred FLUX 3 over Runway Gen-4.5 in 77 percent of comparisons and over Luma Ray 3.2 in 93 percent. The company has not yet published full benchmark results; it says complete benchmarks and methodology will follow alongside broader availability, so these early numbers should be read as the company’s own reported preference rates rather than independent scores.
The model is already being tested by creative platforms including Canva, Burda, Magnific (formerly Freepik), Krea and Picsart. Earlier FLUX models power generative features inside Adobe Photoshop, Picsart and Nous Research’s Hermes Agent, which gives FLUX 3 an existing distribution channel that most new video models lack.
FLUX-mimic: robotics tested on the Audi production line
FLUX-mimic, built with mimic robotics on top of FLUX 3, is designed for general-purpose robotic manipulation: understanding a visual scene, predicting the consequences of an action, and adapting to new tasks with less task-specific data. Depending on task difficulty, the model can be fine-tuned for a specific manipulation task with as little as 30 minutes of robot data, where prior approaches required 30 or more hours.
The model is undergoing testing and deployment with manufacturing companies including Audi. According to Audi’s Production Lab, robots running FLUX-mimic have solved complex soft-body manipulation tasks that were not feasible with conventional robotics, with applications in production and logistics operations.
FLUX 3 availability and open weights
FLUX 3 Video, with optional native audio generation, and FLUX 3 Action are available in early access as of July 23, 2026. Developers and teams can apply for access through Black Forest Labs, with general availability to follow. FLUX 3 Image will roll out in the coming weeks.
Black Forest Labs will also release faster and open-weight versions of FLUX 3 later this year, in line with its practice of publishing open models. Open weights make local, low-latency deployment possible for use cases such as robotic control systems, and let teams fine-tune FLUX 3 on their own data. The company, which employs around 100 people in Freiburg and San Francisco and is valued at 3.25 billion dollars, has raised more than 450 million dollars from investors including a16z, NVIDIA, Salesforce Ventures and Adobe Ventures.
Full details are available in the official announcement of Flux 3 on bfl.ai.
