Black Forest Labs announced FLUX 3 on July 23, 2026, and put its video model into invite-only early access. Unlike every FLUX model before it, the headline feature isn't images — it's video, with sound.
FLUX 3 is designed as one model that generates video, images, and native audio together. Today you can't download it or use it freely; access is limited to teams that apply.
What FLUX 3 can do
Black Forest Labs describes FLUX 3 Video as more than a text-to-clip tool. It handles references, extensions, and editing inside the same model:
| Capability | What it means |
|---|---|
| Text-to-video | Generate a clip from a written prompt |
| Image-to-video | Use an image as a style reference or as the literal first frame |
| Video reference | Feed an existing clip to guide the look of a new one |
| Video + audio extension | Give it footage *and* audio, and it continues both |
| Keyframe control | Pin specific frames; the model fills in the motion between them |
| Smart editing | Stitch single takes into longer, multi-shot sequences |
| Text and animation | Render legible on-screen text and motion-graphic animation |
| Image output | It still produces stills, not just video |
The biggest change is native audio: dialogue, effects, and ambience come out of the same generation pass instead of being layered on afterward, and it supports multiple languages.
Black Forest Labs also points to many visual styles beyond the usual cinematic default.
Early-access testers posting clips on launch day cited first-impression numbers: around 20 seconds of video per prompt, up to 10 reference media (images and clips), and aspect ratios out to 21:9. Black Forest Labs hasn't published formal specs, so treat those as user reports, not guarantees.
What's available now — and when the rest arrives
FLUX 3 isn't one button you can press today. It's rolling out in stages, and only part of it is live.
| Stage | Status (as of July 24, 2026) |
|---|---|
| FLUX 3 Video (+ optional native audio) | Early access now — invite-only |
| FLUX 3 Image | Coming in the following weeks |
| Open weights (FLUX 3 Dev) | Later in 2026 |
Access is granted to teams that apply through Black Forest Labs.
Black Forest Labs says FLUX 3 Video "already leads in early evaluations against frontier video models." That's the lab's own claim about a model still in development — no independent benchmark has been published yet.
How FLUX 3 differs from FLUX.2
FLUX.2 is an image model, and it's still the current one you can use today. FLUX.2 Pro is available now through standard APIs.
FLUX 3 is the new multimodal model, and video is its centerpiece. Its API is expected in the coming weeks.
Need to make video this week?
FLUX 3 can't help yet, but several video models already can.
Sora 2 and Google Veo 3.1 both do text-to-video with synchronized audio, and Kling 3.0 is well suited to motion and image-to-video. All three are available today through a standard API.
For a side-by-side look at the current field, our 2026 AI video model comparison covers Seedance, Kling, Sora, and Veo on quality and cost.
FAQ
When is FLUX 3's release date?
It was announced on July 23, 2026. FLUX 3 Video is in early access now, FLUX 3 Image is expected within weeks, and open weights are slated for later in 2026.
Is FLUX 3 free?
Not as of July 24, 2026. Access is invite-only early access, and no pricing or free tier has been announced.
Can I download FLUX 3?
No. Downloadable open weights (the FLUX 3 Dev version) are planned for later in 2026.
Is FLUX 3 open source?
Not yet. Black Forest Labs says open-weight versions will follow later in 2026; for now only hosted early access exists.
How is FLUX 3 different from FLUX 2?
FLUX.2 is an image model. FLUX 3 is multimodal, and its headline capability is video with native audio.