The short answer: FLUX 3 is Black Forest Labs' multimodal model — image, video with native audio, and robot action prediction, all from a single set of weights. It was announced on July 23, 2026. But as of this writing only the video and action components are open, through gated early access for selected partners. Image generation hasn't shipped yet. If you came here to try it today, that's the part most coverage skips.
Availability at a glance — as of July 27, 2026
- FLUX 3 Video — open, but gated early access (selected partners)
- FLUX 3 Action (robotics) — open, gated early access
- FLUX 3 image generation — not yet released; stated as "coming weeks"
- FLUX 3 Dev (open weights) — planned for later, no public date
What Makes FLUX 3 Different
Most "multimodal" launches are separate models bolted together — one generates the picture, another generates the clip, a third adds sound. FLUX 3 is trained the other way round. Black Forest Labs calls the approach Self-Flow: a unified flow-matching framework where image, video and audio are learned simultaneously so that each modality constrains the others.
The reasoning behind it is the interesting part. Images teach the model spatial structure and composition; video adds temporal dynamics and physics; audio encodes the causal relationship between an event and the sound it makes. Train them together and the model doesn't just render a scene — it has to render one where the motion and the sound agree with each other.
FLUX 3 Video: 20 Seconds with Native Audio
The headline capability is single-generation clips of up to 20 seconds, with audio produced natively by the same model rather than dubbed on afterwards. Native audio is no longer unique — Seedance 2, Sora and Veo all generate sound with the picture — so the real jump here is length: most current video models cap out around 5 to 15 seconds per generation.
For anyone producing short-form content, the practical effect is fewer joins. A 20-second take means a full ad or story beat can land in one generation instead of three stitched clips with a continuity problem in the middle.
Beyond Media: Robot Action Prediction
The genuinely unusual part of FLUX 3 is that the same backbone predicts robot actions. Black Forest Labs built FLUX-mimic with mimic robotics, pairing the FLUX 3 video model with robot learning for dexterous manipulation, and reports that Audi is testing it for production tasks — with some tasks fine-tuned on as little as 30 minutes of robot data.
That's not directly useful if you make TikToks. But it tells you something about the direction: BFL is treating video generation as a world model, not a rendering trick.
How FLUX 3 Compares
An honest comparison is hard right now, because FLUX 3 isn't broadly testable. On stated specs:
| Model | Clip length | Native audio | Availability |
|---|---|---|---|
| FLUX 3 (Black Forest Labs) | Up to 20s | Yes | Gated early access; image side not shipped |
| Seedance (ByteDance) | Up to 15s | Yes | Generally available, including unlimited plans |
| Sora / Veo | Varies | Yes | Generally available, credit-based |
Specs as publicly stated on July 27, 2026. FLUX 3 figures come from Black Forest Labs' launch materials and press coverage rather than hands-on testing, since access is still gated.
How to Get Access — and What to Use Meanwhile
There's no open sign-up yet. Access runs through Black Forest Labs' partner program, and several platforms currently advertising FLUX 3 are either waitlisted or quietly serving FLUX.2 in the meantime. If a tool promises FLUX 3 image generation today, it's worth asking which model is actually running, because the image component hasn't shipped.
If you need to generate today rather than wait: on Creaa.ai you can generate unlimited with GPT Image 2 and Nano Banana 2 for images, and Seedance 2 for video — no credits to count. We're tracking FLUX 3 availability and will update this page when it can be used here.
Keep Reading
FAQ
What is FLUX 3?
Black Forest Labs' multimodal model — images, video with native audio, and robot action prediction from one set of weights, trained together under an approach BFL calls Self-Flow.
When was FLUX 3 released?
Announced July 23, 2026, in a staged rollout: video and action first via gated early access, image generation stated for the following weeks, open weights later.
Can I use FLUX 3 right now?
Not openly. Access is gated to selected partners, and image generation hasn't shipped. Tools claiming FLUX 3 image output today are most likely running FLUX.2.
How long can FLUX 3 videos be?
Up to 20 seconds in a single generation, with audio generated natively by the same model.
Is FLUX 3 free or open source?
Not at launch. An open-weight FLUX 3 Dev backbone is planned for later, but no public date has been given.
Status and specifications reflect Black Forest Labs' public announcements and press coverage as of July 27, 2026, and will change as the rollout continues — we update this page as that happens. Figures for FLUX 3 are as stated by BFL, not hands-on measurements, because access remains gated. FLUX and FLUX 3 are models by Black Forest Labs; Seedance is by ByteDance; Sora is by OpenAI; Veo is by Google. Creaa.ai provides hosted access to the models listed on its pricing page and is not affiliated with any of these providers.