FLUX 3 video enters limited release with 20-second clips and audio

Black Forest Labs launched FLUX 3 Video as part of a broader FLUX 3 model family that can generate images and audio-video clips of up to 20 seconds from one prompt. The company did not disclose pricing, API availability, service commitments or full benchmark methodology, which limits what enterprise buyers can evaluate today.

The Freiburg, Germany-based AI lab says FLUX 3 was trained across images, video, audio and action prediction inside one architecture, rather than by connecting separate models behind a single interface. Black Forest Labs is pitching that approach as a foundation for creative software, simulation, computer use and robotics, a category it calls visual intelligence.

The release is also the company’s first public video generation model. It arrives through four lines: FLUX 3 Video, FLUX 3 Image, FLUX 3 Action and FLUX 3 Dev. FLUX 3 Video and FLUX 3 Action are now in a gated early access program, with applications subject to approval by Black Forest Labs. FLUX 3 Image is expected in the coming weeks, followed by broader availability.

What can FLUX 3 video do?

Black Forest Labs says FLUX 3 Video can produce video with native audio, handle text-to-video and image-to-video generation, use reference clips for video-to-video generation, continue existing audio-video input, create keyframe-controlled transitions and support multilingual dialogue. The company also says individual clips can be chained into longer multi-shot sequences while maintaining visual references such as characters.

The 20-second single-generation length is notable in a market where many production tools still work in shorter clips. Black Forest Labs has not stated the maximum resolution for FLUX 3 Video, while its published evaluations used 720p clips.

The company claims FLUX 3 is strong at facial expression, matching sounds to physical events and multilingual output. It also says the image model improves over earlier FLUX versions on complex prompts and text rendering, including multiple languages. Black Forest Labs has not published image-model benchmarks.

How does FLUX 3 compare with other video models?

Black Forest Labs published preliminary preference results for 10-second, 720p text-to-video clips with audio. It says an early FLUX 3 candidate was preferred over Luma Ray 3.2 in 93% of comparisons, Runway Gen-4.5 in 77%, Grok Imagine Video in 69%, Kling v3 Pro in 60%, Happy Horse v1 in 59%, Happy Horse 1.1 in 57%, and both Seedance 2.0 and Google’s Gemini Omni Flash in 52%.

Those numbers come with a material caveat from Black Forest Labs: they describe an early candidate, not necessarily the model now entering early access. The company says full results and methodology will come later, so buyers cannot yet reproduce the comparisons or judge sample sizes and rater design.

The closest platform comparison in the disclosed results is Google’s Gemini Omni Flash, where Black Forest Labs reported a 52% preference rate. Google’s product is available through the Gemini API at $0.10 per second for generated 720p video, while FLUX 3 pricing has not been announced.

Where does robotics fit into FLUX 3?

Black Forest Labs is extending the same architecture into action prediction through FLUX 3 Action and FLUX-mimic, a model built with Swiss firm Mimic Robotics. The companies say FLUX-mimic uses the FLUX 3 video backbone with Mimic’s robot-learning work for dexterous manipulation.

Black Forest Labs and Mimic Robotics claim the system can be fine-tuned for some manipulation tasks with as little as 30 minutes of robot data, compared with 30 or more hours for prior approaches. The claim is significant for robotics teams if it holds in deployment, but the companies did not provide a standardized benchmark in the launch materials.

Open weights are also delayed. Black Forest Labs says faster and open-weight FLUX 3 versions will arrive later this year, including FLUX 3 Dev, but it has not shared the license, parameter count, quantization plans or hardware requirements. That matters because earlier FLUX Dev releases helped the company gain adoption among developers and creative-tool vendors through local deployment and integration workflows.

Black Forest Labs says FLUX models are already used in products from Adobe Photoshop, Picsart and Nous Research’s Hermes Agent. The company says it has a 100-person team across Freiburg and San Francisco, a $3.25 billion valuation and more than $450 million raised from investors including a16z, Salesforce Ventures, Nvidia, General Catalyst, Adobe Ventures, Figma Ventures, Canva and Deutsche Telekom’s T.Capital.

This story draws on original reporting from VentureBeat.