FLUX 3 — Black Forest Labs' New Video Model, Explained
Announced July 23, 2026. Gated early access. No public API yet.
Black Forest Labs announced FLUX 3 on July 23, 2026 — one multimodal model for image, video, audio and action prediction, built on a method it calls Self Flow, with FLUX 3 Video and FLUX 3 Action behind an application-gated early access program. Here is what BFL officially claims, what nobody has independently tested yet, and what you can actually generate today while you wait.
Official Video
Watch: FLUX 3, as Black Forest Labs Showed It
The announcement video Black Forest Labs published on July 23, 2026. Every clip in it was rendered by BFL — none of it was generated on ImagineToVideo, and none of it has been independently reproduced.
FLUX 3 at a Glance — Official Claims, Not Yet Independently Tested
Up to 20 Seconds With Native Audio (Claimed)
Black Forest Labs states FLUX 3 Video generates up to 20 seconds in a single pass with natively generated audio, with its published evaluations run at 720p. ⚠️ Official claim — the model is not publicly released, so no third-party test has verified it.
One Model, Many Video Tasks (Claimed)
BFL lists text-to-video, image-to-video, video-to-video, video-audio continuation, keyframe-to-video, multilingual dialogue and agentic chaining for longer sequences, built on a method it calls Self Flow. ⚠️ This is the official capability list, not an independently verified one.
Gated Early Access — Not Publicly Available
FLUX 3 Video and FLUX 3 Action are behind an application-gated early access program. FLUX 3 Image is described as coming in the following weeks, and the open-weight FLUX 3 Dev later this year. As of this page, there is no public API and no published pricing — the BFL pricing page still lists FLUX.2 and FLUX.1 only.
What You Can Generate Right Now
No application, no waitlist: Wan 3.0 (3–30 seconds, 480p/720p/1080p, native audio), Seedance 2.5 (4–30 seconds, 480p/720p, native audio, plus video-to-video), Kling V3 and Kling V3 Omni (3–15 seconds, 720p/1080p, native audio), VEO 3.1 and VEO 3.1 Fast (4/6/8 seconds, 720p/1080p, native audio), and Seedance 2.0 (4–15 seconds, 480p/720p/1080p, native audio). All live on ImagineToVideo today.
What FLUX 3 Is Aimed At — And What Covers It Today
Dialogue-Driven Ad Spots
FLUX 3 claims native audio and multilingual dialogue in one pass. Today, VEO 3.1 already turns one prompt into picture and sound together at 4, 6 or 8 seconds — the right length for a paid social cutdown.
Longer Takes & Multi-Shot Sequences
FLUX 3 claims 20 seconds in a single generation (⚠️ untested). Today, Seedance 2.5 already runs up to 30 seconds in a single pass at 480p/720p — longer than the FLUX 3 claim — while Kling V3 covers 3–15 seconds per take at up to 1080p, and multi-scene chaining keeps a character consistent across several connected shots.
Video-to-Video & Continuation
BFL lists video-to-video and video-audio continuation among FLUX 3's modes. Today, Seedance 2.5 already does video-to-video on ImagineToVideo — feed in existing footage, a reference image or a last frame, and get up to 30 seconds back.
Robotics & Embodied AI
FLUX 3 Action, developed with mimic robotics, extends the same backbone to predicting robot actions — and sits inside the same gated early access. That work is outside consumer video generation; follow the BFL blog for it, not us.
How to Work With FLUX 3 Today
One honest path: get in line for FLUX 3, and keep shipping video in the meantime
Check FLUX 3's Real Status First
Open the Black Forest Labs FLUX 3 announcement and submit an early access application for FLUX 3 Video or FLUX 3 Action. Until BFL approves you there is no endpoint to call — FLUX 3 Image is still described as weeks away, and FLUX 3 Dev weights later this year.
Write a Model-Portable Prompt
Put the shot description, spoken lines and ambient sound into a single prompt. Every audio-native video model — the one FLUX 3 claims to be, and the ones running today — reads prompts this way, so the same prompt carries over when FLUX 3 opens up.
Pick a Model That Ships Audio Today
For spoken dialogue choose VEO 3.1. For camera movement and 15-second single takes choose Kling V3. For a single take of up to 30 seconds, or to transform footage you already have, use Seedance 2.5 — it is also the model that runs video-to-video. All of them sit in one panel at /video/text-to-video.
Draft Cheap, Then Scale Up
Lock motion and composition with a 720p draft, then re-run the winning take at 1080p. For longer narratives, chain shots in /video/multi-scene. When FLUX 3 becomes available, re-run the same prompt and A/B it against what you already shipped.
FLUX 3 FAQ
What is FLUX 3?
FLUX 3 is a multimodal model family announced by Black Forest Labs on July 23, 2026, trained jointly on images, video and audio and extendable to action prediction. It spans FLUX 3 Video, FLUX 3 Action, FLUX 3 Image and an open-weight FLUX 3 Dev, built on a method BFL calls Self Flow. Black Forest Labs was previously best known for its FLUX image models.
Can I use FLUX 3 right now?
Not freely. FLUX 3 Video and FLUX 3 Action are in an application-gated early access program — anyone can apply, but BFL has to approve you. FLUX 3 Image is described as arriving in the following weeks, and the open-weight FLUX 3 Dev later this year. The named early partners are Canva, Burda, Magnific (formerly Freepik), Krea and Picsart.
How long can FLUX 3 videos be, and does FLUX 3 generate audio?
Black Forest Labs states FLUX 3 Video produces up to 20 seconds in a single generation with native audio, and lists keyframe-to-video, video-audio continuation and multilingual dialogue among its modes; the published evaluations were run at 720p. ⚠️ These are BFL's own stated specifications. The model has not been publicly released, so there is no third-party measurement to confirm them.
Is FLUX 3 better than Kling, VEO, Runway or Luma?
The win rates Black Forest Labs published are 69% against Grok, 60% against Kling v3 Pro, 77% against Runway Gen-4.5 and 93% against Luma Ray 3.2. ⚠️ These are vendor self-reported preference results, not independent blind tests — BFL has not published its evaluation protocol, sample size or prompt set. Treat them as a vendor claim and wait for hands-on comparisons once FLUX 3 is generally available.
Does FLUX 3 have a public API or published pricing?
Not as of this page. Black Forest Labs says API access, private model weights and the open-weight FLUX 3 Dev are planned for later this year, but nothing is live yet. The BFL pricing page still covers FLUX.2 and FLUX.1 models only — there is no FLUX 3 SKU and no per-second or per-generation rate, so any FLUX 3 price quoted today has no official source behind it.
What can I use instead of FLUX 3 today, and will ImagineToVideo add it?
Available today with no approval step: Wan 3.0 (3–30 seconds, 480p/720p/1080p), Seedance 2.5 (4–30 seconds, 480p/720p, video-to-video included), Kling V3 and Kling V3 Omni (3–15 seconds, 720p/1080p), VEO 3.1 and VEO 3.1 Fast (4/6/8 seconds, 720p/1080p), and Seedance 2.0 (4–15 seconds, 480p/720p/1080p) and Seedance 2.0 Fast (4–15 seconds, 480p/720p) — all with native audio. On FLUX 3 itself: we plan to evaluate integration once the official API and pricing are public, benchmark the real specifications against what has been claimed, and then price it in credits. The timing depends on when Black Forest Labs opens the API and publishes pricing, so we are not committing to a date. Our full reasoning is in our FLUX 3 early access write-up on the blog.
Don't Wait on a Waitlist to Ship Video
FLUX 3 is behind an application gate with no public API and no published pricing. Wan 3.0, Seedance 2.5, Kling V3 and VEO 3.1 are one click away — native audio, up to 30 seconds per clip on Seedance 2.5, and 720p and 1080p on Kling V3 and VEO 3.1.