One backbone, many modalities
FLUX 3 jointly learns image, video and audio so motion, appearance and sound stay causally consistent — footsteps match impact, lips track speech, physics stay believable.
New · FLUX 3 · Black Forest LabsFLUX 3 AI Video GeneratorVideo + Native Audio · Free to Start
On July 23, 2026 Black Forest Labs introduced FLUX 3 — not “another image model,” but a unified multimodal foundation trained jointly on image, video and audio (Self-Flow). FLUX 3 Video targets up to ~20-second clips with optional native audio, plus text/image/video conditioning paths. Kivide is your creator studio to explore the FLUX family alongside Seedance, Kling and Wan — free credits, clean exports.
FLUX 3Self-FlowNative audio~20s videoBFLEarly Access era
Duration (seconds)5s
AI Video
Why FLUX 3 matters for video creators
Up to ~20s with native audio
Generate longer takes with optional synchronized audio. Early public evals often cite 10s @ 720p with audio while the capability ceiling is described around 20 seconds.
Rich conditioning paths
Text-to-video, image-to-video, video-to-video, keyframe transitions and multi-clip agentic chaining — built for multi-shot storytelling, not single demo loops.
Kivide multi-model workspace
Use FLUX for art-directed looks, Seedance for cinematic motion, Kling/Wan for speed — one account, transparent credits, watermark-free downloads.
Create with FLUX on Kivide in 3 steps
Write a cinematic prompt
Include subject, camera, lighting, mood — and if you want sound, describe ambience or dialogue briefly.
Choose FLUX in the studio
Select the FLUX engine for art-directed output. As FLUX 3 Video capacity is enabled on our stack, generations route to the newest available build automatically.
Generate & download
Preview, then export a clean file. Free credits on signup — no credit card required to start.
FLUX 3 vs prior FLUX (direction)
| Capability | Kivide | Typical |
|---|---|---|
| Core focus | Image + video + audio (+ action path) | FLUX.1 / 2 image-first |
| Video length | Up to ~20s + native audio (3) | N/A or non-core on prior FLUX |
| Architecture | Self-Flow multimodal foundation | Image generation oriented |
| Try on Kivide | Free credits · multi-model | Single-vendor waitlists |
Model matrix
Every model intro on Kivide
FLUX 3 FAQ
FLUX 3 is Black Forest Labs' multimodal frontier model announced July 23, 2026. It is trained jointly on images, video and audio under a Self-Flow / flow-matching design, and can extend toward action prediction (e.g. FLUX-mimic with robotics partners). Product lines include FLUX 3 Video, Image, Action and a later open FLUX 3 Dev backbone.