Black Forest Labs ships FLUX 3, its first video model with native audio output

Black Forest Labs launched FLUX 3 today, marking the Freiburg-based AI lab's first entry into video generation. The model produces video clips up to 20 seconds with synchronized native audio output — both generated jointly from a single prompt using a unified architecture, rather than routing separate image and audio models behind a common API.

The distinction matters. Most video-generation products today are assemblies: a text encoder, a video diffusion model, and a separate audio model stitched together. FLUX 3 is built on Self-Flow, BFL's architecture for training across images, video, and audio simultaneously, so the model develops a shared representation of spatial structure, motion physics, and acoustic causality. According to the company, that mutual training signal is why the same weights can power both a photorealistic film clip and a manipulation task in a robotic arm.

What ships today

FLUX 3 launches across four product lines. FLUX 3 Video and FLUX 3 Action enter a gated Early Access program today — available to anyone who applies, pending BFL approval. FLUX 3 Image will roll out publicly over the coming weeks. FLUX 3 Dev, an open-weight version of the multimodal backbone, is committed for later this year but has no confirmed date.

FLUX 3 Video supports text-to-video, image-to-video (animation or visual reference), video-to-video style transfer, keyframe-controlled transitions, multilingual dialogue generation, and agentic chaining of individual clips into longer sequences. All video outputs include native audio. FLUX 3 Action extends the same architecture to robotic perception and physical control — a notable ambition shift for a lab that built its name on consumer image generation.

Benchmark results, with caveats

BFL published preliminary preference evaluations for FLUX 3 Video against seven competitors on 10-second, 720p text-to-video clips. FLUX 3 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%, and Seedance 2.0 and Google Gemini Omni Flash each at 52%.

BFL labels these results as a "preliminary evaluation of an early candidate" — meaning the benchmarked checkpoint is not the model entering early access today. The company says the shipping model may perform better, and full methodology and benchmark tables are promised at general availability. The two tightest comparisons — 52% against both Seedance 2.0 (whose international rollout has been suspended due to copyright litigation from major Hollywood studios) and Gemini Omni Flash — are the most meaningful for Western enterprise buyers.

What is missing

There is no public API at launch and no pricing. Enterprise buyers cannot yet calculate cost of ownership or run independent evaluations. BFL has not released downloadable weights, no open-source license is attached, and production service-level commitments have not been published. FLUX 3 Dev — the open-weight release the developer community has been anticipating — is last in BFL's rollout sequence, arriving after Image reaches general availability.

That sequencing is a departure from the FLUX 1.x releases, where open weights arrived alongside or soon after commercial access and played a central role in the architecture's adoption across third-party applications and platforms. BFL says the FLUX 3 Dev commitment covers "video, audio and image generation and action prediction" — a considerably broader scope than any previous FLUX Dev release.

Why it matters

FLUX 1.x became the default open-weight image generation backbone for a large share of consumer and developer AI products in 2024 and 2025. A credible multimodal successor that eventually ships open weights would extend that footprint into video and robotics. Whether FLUX 3 achieves that position depends on what the open-weight release looks like when it arrives — and on BFL's ability to close the gap against Sora, Veo 3, and Kling between now and general availability.

Applications for FLUX 3 Video Early Access are open at bfl.ai. FLUX 3 Image general availability is expected in the coming weeks, as first reported by VentureBeat.