Flux 3: Black Forest Labs Just Became a Video Company

Short version: Black Forest Labs, the lab behind the Flux image models, announced Flux 3 on 23 July 2026. It is not a better image model. It is one model trained jointly on images, video and audio, generating clips up to 20 seconds with sound that actually lines up. There is also a robotics variant. The image lab quietly became a video lab.

Quick flag: Leaxor is mine, and it now runs Flux 3, so read the recommendation at the bottom with that in mind. Every price below is fal's published rate or our own credit cost, both checkable.

What Black Forest Labs actually shipped

Flux 1 and Flux 2 made pictures. That was the whole product, and they were good enough at it that Flux 2 Pro is the default image model on this platform.

Flux 3 is a different kind of thing. It is a single model trained jointly across images, video and audio, then extended to predict robot actions. VentureBeat's launch coverage put the headline capability plainly: text-to-video up to 20 seconds carrying native, in-sync audio.

The framing the lab used is "visual intelligence" rather than image generation. That sounds like positioning until you notice the robotics part, which is where it stops sounding like marketing.

Why joint audio is the actual news

Almost every video model hands you a silent clip. You then find music, cut it to length, add footsteps and room tone, and hope none of it drifts against the picture.

Flux 3 generates the audio with the frames rather than after them. Dialogue lands on the mouth. A door closing sounds when the door closes. That is not a small convenience, it removes a whole stage of post-production for anyone making short-form video.

Sora 2 had this before it was discontinued, and its loss took the capability off the table for most people. Flux 3 puts it back, which matters more than the spec sheet suggests.

The robotics part, briefly

Flux 3 Action predicts robot actions using the same machinery that predicts the next video frame, and MarkTechPost reported a variant already being tested on Audi production lines.

No creator will ever touch this. It is worth one paragraph because it explains the architecture: if you believe predicting the next video frame and predicting the next physical action are the same problem, you build one model for both. That belief is why the video half exists at all.

What Flux 3 video costs

Priced per render on Leaxor, close to what the model costs us to run:

Tier	5 seconds	8 seconds	10 seconds
Flux 3 draft, 720p	4 credits ($0.40)	6 credits ($0.60)	7 credits ($0.70)
Flux 3, 720p	10 credits ($1.00)	16 credits ($1.60)	20 credits ($2.00)
Flux 3, 1080p	17 credits ($1.70)	27 credits ($2.70)	34 credits ($3.40)

Credits are $0.10 each, minimum purchase 50 credits, no subscription and no expiry.

One habit worth forming regardless of price: the draft tier is a third of the cost and exists so you can be wrong cheaply. Generate in draft, pick the shot that works, re-render that one at full quality. Doing it the other way round is how a session gets expensive.

The release was staged, and that is unusual

Black Forest Labs built its reputation on open weights. Flux 3 did not arrive that way.

Video and Action opened first, in gated early access with an application and manual approval. The image half followed. The open-weight Dev release was scheduled last.

Reading that as a change of direction is reasonable. A lab that ships its frontier capability behind an approval queue is behaving like a commercial video lab, not like the open-weights outfit people knew. Whether the Dev release lands on time is the thing to watch.

Where Flux 3 fits, and where it does not

Use it when the clip needs sound. That is the one job where nothing else here competes properly.

Use it when you need length. Twenty seconds is generous. Most models cap between 8 and 15.

Do not use it for exploration. Full-quality renders are the dearest thing on the shelf. That is what the draft tier is for, and what Wan 3.0 at 30 cents a clip is for.

Do not use it for finished videos. Like every clip generator, it returns footage. Script, narration, captions and continuity across shots are still yours to solve.

What it costs to try

Flux 3, its draft tier, and the image-to-video and first-and-last-frame variants are all live in Leaxor's clip generator, priced per render with credits that do not expire. There is no subscription and no free tier, so the honest entry cost is a $5 minimum top-up, which is twelve draft clips.

If what you want is a finished narrated video rather than footage to edit, that is a different mode and a different price. And if you only want cheap motion to see whether an idea works, start with the draft tier or with Wan 3.0 and come back to Flux 3 when the shot matters. That ordering will save you more money than any discount.

The wider wave this landed in is covered in the August 2026 model roundup, and the per-second cost of every model on the shelf is laid out in the pricing breakdown.

Reading this with an AI assistant? Ask it to summarize this in ChatGPT, Claude, or Perplexity.