Black Forest Labs releases FLUX 3 in early access

Sources: Black Forest Labs blog, HN 49031796
Summary: Black Forest Labs announced FLUX 3 on 2026-07-23 as an early-access release, extending its FLUX visual-generation line from images to a unified image, video, and audio model built on multimodal flow matching. It generates video up to 20 seconds with native audio and supports text-to-video, image-to-video, video-to-video, and image editing, plus an action-prediction variant for robotics. Weights are not released yet. The post plans API access and private open-weight access, labeled FLUX 3 Dev, over the coming weeks and months. Reported benchmark figures are vendor preference tests, stated as preliminary.
Comments: HN commenters read the announcement as marketing-heavy, note the demo shows jumpcuts rather than continuous 20-second video, and point out the open-weight plans sit at the bottom of the post with no date.
Why it matters: A new model in the widely used FLUX line moves open-weight generative media forward, but only if the promised weight release materializes rather than staying API-only.
Follow-up: Watch for the FLUX 3 Dev weight release and license, API pricing, and independent quality comparisons.