Black Forest Labs made FLUX 3 Video generally available on 4 August through its API and selected partners. The launch turns the earlier multimodal research programme into a working video product with text-to-video, image animation, keyframes, continuation and sound generated with the pictures.
Black Forest Labs Watch analysis
What happened
The service generates clips from five to 20 seconds at 720p, with 1080p available through upscaling. Users can begin from text, animate an opening image, specify an end frame, pin several timed keyframes or continue an existing video and its audio across the join.
Native sound covers dialogue, effects and ambience in the same request. The release also includes multiple-shot generation, multilingual speech and a draft mode intended to preserve the approved composition and motion when a preview is rendered at full quality. Those are product claims from Black Forest Labs, not independent guarantees across every subject or workflow.
Why it matters
Professional adoption depends less on a single spectacular clip than on whether a creator can revise, extend and reproduce a shot without starting again. Keyframes and continuation attack that problem directly by giving people anchors at the beginning, middle or end of the generation.
Synchronous audio also removes a separate assembly step, but it creates another failure surface. Speech, lip movement, impacts and ambience all need to agree with the pictures. Company preference studies are useful release evidence, not a substitute for independent testing of temporal consistency, editability and cost.
What to watch
- Independent tests of keyframe fidelity, continuation seams and long-shot consistency.
- Whether native dialogue and effects remain synchronised across languages and camera cuts.
- Real production cost, queue latency and iteration speed compared with assembling specialised tools.
Our read
This launch is interesting because it packages several awkward production handoffs into one controllable route. The test is whether those controls survive revision rather than merely behaving for the launch reel.
A 20-second clip is plenty of time for a cufflink to migrate, a sentence to lose its mouth and gravity to nip out for a cigarette. Reliable control is the product.
Discussion spark: Which matters most for professional video work: keyframes, continuation, native audio, temporal consistency or iteration cost?
Sources and evidence
- FLUX 3 Video, Part 1: Generation (4 August 2026, 00:00 UTC)
- FLUX 3 Video documentation (4 August 2026, 00:00 UTC)
not affiliated with or endorsed by Black Forest Labs