Black Forest Labs unveiled FLUX 3 on July 23 as one model for images, video, audio and robot actions, and opened FLUX 3 Video behind an application gate, with no API access on day one. That gate is gone. The company's model page now lists FLUX 3 Video as available through its dashboard and its API on a pay-as-you-go basis, with a public price list, which is the moment a research launch turns into a product that any developer can put in a pipeline.
The video model generates clips of up to 20 seconds in a single pass, in HD (up to one megapixel per frame) or FHD (up to two megapixels per frame), with native audio: multilingual dialogue, speech synchronised to the picture, sound effects and ambient sound, included at no extra charge. The API exposes four ways in. Text-to-video and image-to-video are the obvious two. Keyframe-to-video takes up to ten stills and animates the transitions between them, which is how a storyboard becomes a sequence. Video continuation extends an existing clip by five to fifteen seconds per pass, so longer pieces are built by chaining. A draft mode renders a faster, cheaper HD preview that can be enhanced to full FHD once the shot is approved, which is closer to how a studio actually works than the generate-and-pray loop of earlier tools.
Prices are per second of output. Text and image to video cost 0.06 dollars per second in draft HD, 0.17 in standard HD and 0.29 in standard FHD. Continuing a video costs more, 0.12, 0.41 and 0.53 dollars per second on the same three tiers. Black Forest Labs' own example puts a five-second standard HD clip from text at 0.85 dollars. For comparison, the 20-second ceiling at FHD works out to just under six dollars before any retries.
The launch pitch was that a model which only learns from images can only make images. "You can't cheat reality," chief executive Robin Rombach told VentureBeat in July, and the company has been feeding the same backbone to a robotics variant, FLUX-mimic, that Audi says is running on its production lines for soft-body manipulation. The rest of the family is still staged: FLUX 3 Image, FLUX 3 Action and the open-weight FLUX 3 Dev backbone are listed as separate parts of the rollout, and the image model has not reached the public API. Canva, Burda, Magnific, Krea and Picsart were named as testing partners at launch.
Zubnet added FLUX 3 Video to its platform today, with the same three per-second tiers, the draft mode, keyframes and continuation, priced by the real length of the clip that comes back rather than by the length requested.