New Video AI FLUX 3 - multimodal model with native audio generation
[11:51 Sun,26.July 2026 � by Thomas Richter] ���
It took a while, but now it is here: the first video model from the Freiburg-based AI startup Black Forest Labs. Almost exactly 2 years ago, it introduced its image AI Flux, which has since been continuously improved, integrated into numerous tools, and has even impressed Hollywood legend Martin Scorsese.
FLUX 3 Multimodal
Black Forest Labs has kept us waiting for a long time, but is now delivering its first multimodal model capable of generating not only images, but also videos complete with matching audio (both sound effects and speech). FLUX 3 is designed to be state-of-the-art and competitive with current top models – also thanks to a comprehensive understanding of the physical world, including the rules of causal interaction between objects and how these are described by text. The demo impressively shows how well FLUX 3 can render things like fluid behavior, the movements of animals or even fingers, and wild camera movements – though notably absent are close-ups of people, their faces, and their movements.
What FLUX 3 Can Do
FLUX 3 generates 720p video clips up to 20 seconds long with native, synchronous audio. It supports text-to-video, image-to-video (animation or visual reference), video-to-video transfer, keyframe-controlled transitions, and, of course, rendering in various visual styles. Multiple clips can be connected into longer sequences via agentic chaining. According to BFL, this makes it possible to create multi-minute sequences with consistent characters using reference images, essentially amounting to short films. Black Forest Labs also presents a benchmark test indicating that Flux managed to keep pace with current top models like Google Omni Flash and Seedance 2.0 – it remains to be seen whether independent tests will confirm this.
FLUX 3 Comparison with other models
What FLUX 3 Can Do in Detail
- Text-to-video generation with native audio - Image-to-video generation, either as a continuation of an initial frame ("animation") or using images as visual references. - Video-to-video generation based on a reference clip, transferring core elements of an original video – such as the same character – into a new scene or context. - Generative video-audio continuation based on inputted video and audio. - Keyframe-to-video generation for controlled transitions between defined moments. - Multilingual dialogues. - A wide range of visual styles and aspect ratios that go far beyond conventional cinematic outputs. - Agent-based chaining of individual clips into longer sequences with multiple shots. - High stylistic diversity – FLUX 3 Video effortlessly masters a range from authentic camcorder footage to animations and cinematic productions. - Powerful typography generation and animated designs.
FLUX 3
FLUX 3 Image will follow in early access over the coming weeks – BFL promises better prompt processing and multilingual text rendering compared to FLUX.2. At the same time, BFL has developed FLUX-mimic in collaboration with mimic robotics, an action model for grasping manipulation that, according to the manufacturer, is already being tested in production use at Audi.
Availability and Rollout
FLUX 3 Video is available immediately as gated early access application required. FLUX 3 Action, the variant specialized in action prediction, and FLUX-mimic, the video-action model for robotics based upon it, are now available to selected research and industry partners. FLUX 3 Image is scheduled to follow in the coming weeks, and an open-weight release as FLUX 3 Dev is announced for later in 2026.
