ICYMI: Black Forest Labs opens FLUX 3 Video early access

Black Forest Labs, the company behind the widely used FLUX family of image generation models, has begun accepting early access applications for FLUX 3 Video. The new model supports clips of up to 20 seconds in length and - notably - generates audio natively alongside the video, rather than relying on a separate audio pipeline bolted on after the fact.

Native audio generation in a video model is still relatively uncommon. Most video AI tools either skip audio entirely or handle it as a post-processing step, so integrating it into the core generation process positions FLUX 3 Video closer to a small group of models pushing toward fully unified audiovisual output. How well the audio quality holds up across varied prompts will be one of the key things to watch as early access users begin sharing results.

Black Forest Labs has also outlined plans for several related releases. These include an image model update under the FLUX 3 name, an "action" model variant - likely oriented toward motion-heavy or physically dynamic scenes - and an open-weight release. The open-weight commitment in particular continues a pattern the company established with earlier FLUX models, which became popular in the self-hosted and research communities precisely because weights were made publicly available.

FLUX 3 Video enters a competitive field that includes Runway, Kling, and Google's Veo, among others. Black Forest Labs built its reputation on strong image fidelity and prompt adherence with its earlier FLUX.1 models, and it will be worth seeing whether those qualities carry over into the video domain. Early access is currently limited, but the phased rollout - with open-weight models to follow - suggests the company intends FLUX 3 Video to reach a broad audience over time.