Black Forest Labs opens FLUX 3 Video API with native audio and draft rendering

Black Forest Labs, co-founded by CEO Robin Rombach, opened FLUX 3 Video to general API access on August 4th, enabling developers to generate video and synchronized audio from text, images, keyframes, or existing clips. The model supports clips up to 20 seconds in HD, with Full HD output via upscaling, and includes a Draft mode costing $0.06 per second for faster previews. Black Forest Labs claims human raters preferred FLUX 3 in internal evaluations, tying ByteDance's Seedance 2.0 in image-to-video generation and beating other tested systems.

Robin Rombach @robrombach https://x.com/robrombach , co-founder and CEO of Black Forest Labs, opened FLUX 3 Video to general API access on August 4th, giving developers a new model for generating video and synchronized audio from text, images, keyframes or an existing clip. https://x.com/bfl ai/status/2084693191484469305 https://x.com/bfl ai/status/2084693191484469305 Black Forest Labs announced the release in a thread on X https://x.com/bfl ai/status/2084693191484469305 and detailed the rollout in a launch post https://bfl.ai/blog/flux-3-video . The initial version is available through the Black Forest Labs API and selected distribution partners. It generates clips up to 20 seconds long in HD, with Full HD output produced through upscaling. Rombach built Black Forest Labs with researchers including Andreas Blattmann and Patrick Esser after their work on latent diffusion helped produce Stable Diffusion, the open image model that established their standing in generative AI. Andreessen Horowitz described the founders https://a16z.com/announcement/investing-in-black-forest-labs/ as original co-creators of the Stable Diffusion model series when it backed Black Forest Labs in 2024. Black Forest Labs operates from Freiburg, Germany, and San Francisco. FLUX 3 Video is Black Forest Labs' first broadly available move beyond the image-generation market where the FLUX name gained adoption. The FLUX 3 documentation https://docs.bfl.ai/flux 3 lists three launch modes: text-to-video, image-to-video and video continuation. Image-to-video requests can pin multiple images as frames, while video continuation accepts up to four seconds of existing video and audio before generating what follows. The model can create multiple shots and camera angles in one generation. Black Forest Labs also says it can render text inside scenes, generate ambient sound and effects, and produce dialogue with lip-sync in languages including English, Chinese, Spanish, French, German, Japanese, Portuguese, Russian, Italian, Indonesian, Turkish, Hindi and Punjabi. The documentation describes FLUX 3 as a preview model even as Black Forest Labs calls the initial video-generation release generally available. Video editing and a broader reference system using images and videos remain on the roadmap. The current API supports clips at 24 frames per second, including Full HD output at 1920 by 1088 for a 16:9 aspect ratio. Draft first, render later Black Forest Labs is pairing the model with a Draft mode designed to lower the cost of testing prompts. A draft produces a faster preview, and a user can then submit that result for a full-quality render that preserves the subjects, composition and motion of the approved version. Black Forest Labs says Draft mode costs $0.06 per second, putting a 20-second preview at $1.20. The public pricing calculator https://bfl.ai/pricing separately displayed a rate of $0.17 per second for its selected HD configuration on launch day, or $0.85 for five seconds. That workflow targets a persistent expense in generative video: creators often pay for several complete generations before finding a usable direction. The draft-to-render system also gives Black Forest Labs a product distinction beyond output quality. A low-cost preview that reliably carries its composition into the final clip can make FLUX 3 easier to place inside advertising, storyboarding and other production pipelines where repeatability matters as much as an impressive sample. Black Forest Labs' benchmark claims remain internal Black Forest Labs says human raters preferred FLUX 3 in its internal text-to-video and image-to-video evaluations. The August 4th launch post claims the model tied ByteDance's Seedance 2.0 in image-to-video generation and beat the other systems Black Forest Labs tested. The July 23rd early-access announcement reported preference rates of 77% against Runway Gen-4.5 and 93% against Luma Ray 3.2. Those results are company-run evaluations rather than independent benchmarks. Black Forest Labs also said the evaluation harness and model were still under development during early access, limiting direct conclusions about how the release compares across prices, generation times and production workloads. FLUX 3 first appeared in early access on July 23rd https://bfl.ai/blog/flux-3 , when Black Forest Labs presented it as one multimodal foundation model trained jointly across images, video and audio. The same backbone is being adapted for action prediction through FLUX-mimic, a robotics model developed with mimic robotics and tested on manufacturing tasks at Audi. That architecture shows where Rombach is taking Black Forest Labs. Video generation is the first commercial release from a model intended to span media creation, perception and robot control. Black Forest Labs says image generation and editing will follow, along with FLUX 3 Dev, a planned open-weight version of the multimodal backbone. The expansion is backed by substantial capital. In December 2025, Black Forest Labs raised a $300 million Series B https://bfl.ai/blog/our-300m-series-b at a $3.25 billion post-money valuation, co-led by Salesforce Ventures and AMP. Participants included Temasek, Bain Capital Ventures, Air Street Capital, Visionaries Club, Canva and Figma Ventures, alongside existing backers including a16z, Nvidia, Northzone, Creandum, Earlybird, BroadLight Capital and General Catalyst. FLUX 3 Video puts that funding behind a direct challenge to Runway, Luma, Kling, Seedance and Google's video models. Black Forest Labs enters with researchers who helped define open image generation, an established developer distribution strategy and a model designed to combine picture, motion and sound in one request. The August 4th release begins the harder test: whether that research pedigree can translate into recurring API workloads for video.