Black Forest Labs Unveils FLUX 3: Multimodal Model Generates Images, 20‑Second Video with Audio in Early Access

FLUX 3 Launch Details

Black Forest Labs (BFL) announced FLUX 3, a multimodal model that creates images, 20‑second video clips with native audio, and predicts robotic actions from a single textual prompt. Joint training across vision, audio and motion sets it apart from patchwork solutions, positioning it as a true “visual intelligence” platform. The rollout begins with a gated Early Access program for FLUX 3 Video and FLUX 3 Action; applicants must await approval before any API exposure.

“You can’t cheat reality – a model limited to still frames will never capture motion,” said CEO Robin Rombach.

Preliminary head‑to‑head tests show ↑ 93% preference over Luma Ray 3.2 and 77% over Runway Gen‑4.5 on 10‑second, 720p clips, though BFL cautions these figures stem from an early candidate. No pricing, SLA or open‑weight release has been disclosed; the company promises downloadable checkpoints later in the year under the forthcoming FLUX 3 Dev program. Enterprise interest is evident – Canva, Adobe and Picsart are already piloting the technology – but the lack of transparent cost metrics hampers budgeting. For context, Google’s Gemini Omni Flash charges $0.10 per second of 720p video, according to Reuters. BFL’s valuation sits at $3.25 billion after a $450 million raise from a16z, Nvidia and others, underscoring the market’s appetite for unified generative models.

Analysis by Julian Reed (Consumer Electronics Expert).