Alibaba Next-Gen Video Model: What's Confirmed So Far

Thirty seconds was supposed to be the finish line. When Wan 3.0 went public in August with clips up to half a minute, sound included, it looked like Alibaba had answered the length question for a while. Then, at its cloud conference in September, Alibaba said the next one was already in training.

The Alibaba next-gen video model is due in November, according to coverage of that event, and it's aimed at something harder than length: whole stories. Alibaba's next-gen video model doesn't have a public name yet, and November is now well underway with no release. Here's exactly what has been said, what hasn't, and how to get your Wan workflow ready on each::labs before it lands.

What Alibaba Has Actually Said

Almost everything we know traces back to September 22, 2026, and to reports from Alibaba's Yunqi (Apsara) conference. QbitAI, writing from material provided by Alibaba, reported that a brand-new next-generation video model "will also be released in November". Shanghai Securities News carried the same sentence and attributed it to an unnamed person connected to Alibaba. Caiwen, also reporting that day, repeated the November timing.

The same coverage describes the direction in four words: longer, more controllable, more complete and smarter. Over a longer stretch of time, the new model should keep things consistent. Creators should be able to control characters, scenes and camera precisely. And the model should have what the reports call director-level creative thinking, moving from generating single shots to understanding a complete narrative. QbitAI added that the model was still in training, with stronger generation, control and comprehension.

That's the whole list. No name, no spec sheet, no demo clip, no date more precise than "November". We checked Alibaba Cloud Model Studio's list of newly released models and the Wan team's Hugging Face page on November 9, and neither shows a new video model yet.

What We Don't Know Yet

It's worth being blunt about the gaps, because rumors fill them fast. Nobody official has said whether Alibaba's next-gen video model carries the Wan name, and searches for a "Wan 3.5" or "Wan 4" turn up nothing from Alibaba. There's no published maximum length, resolution, frame rate or audio spec. There's no word on whether it ships as open weights, like earlier Wan releases, or as a hosted model, like Wan 3.0.

The November date also comes with a soft edge. One of the three reports attributes it to an unnamed source rather than a named executive, and plans move. We'd treat November as Alibaba's stated target, not a promise, and we'll update this post the moment there's an official release note.

Where Wan 3.0 Leaves Off

To read the November claims properly, look at what Alibaba already ships. Wan 3.0 is live on each::labs today, and Alibaba describes it as an all-in-one model covering text-to-video, image-to-video with first frame or first and last frame, and reference-to-video in a single unified model, with audio generation. Clips run from 2 to 30 seconds, or auto, in 480P, 720P or 1080P. A faster Prime variant offers the same capabilities with quicker generation.

Its reference mode already pushes toward "the brief, not a summary". Wan 3.0 Reference to Video takes up to 10 reference images, up to 5 reference videos and up to 5 audio clips in one request, plus one reference file or one public web page. We covered how to feed it in our Wan 3.0 spotlight.

So the jump Alibaba is describing isn't from short clips to long ones. Wan 3.0 already does half a minute. It's from one good shot to a sequence that holds together as a story.

Director-Level Thinking: Why It's the Real Headline

Every video model this year can make a beautiful shot. Ask for the next shot, and the trouble starts. The jacket changes color, the room rearranges itself, the camera forgets where it was. Teams solve that today by hand: generate shot by shot, pin references, stitch in an editor, and regenerate whatever broke.

"From single shots to understanding a complete narrative" is a claim about that stitching step. A model that understands a story would plan coverage the way a director does, deciding where the wide shot goes and where the close-up lands, and keep characters and places consistent across the cuts. If Alibaba's next-gen video model delivers even part of that, the unit of work shifts from the clip to the scene.

That's also the hardest claim to judge from a sentence. We'll want to see how many shots it holds, how it handles a character who leaves and comes back, and how much of the camera you can still direct. Those are the tests we'll run on day one.

Precise Control: Characters, Scenes, Camera

The second promise is control, and it's the one that matters most in production. Alibaba's reported wording names three things: characters, scenes and camera. Those map neatly onto how teams already work with Wan 3.0, which is a good sign that the new model extends a workflow rather than replacing it.

Characters are the reference images you lock for a presenter or a mascot. Scenes are the locations and set dressing you want to persist. Camera is the motion language in your prompt or in a reference clip. In Wan 3.0 you control all three through references and careful prompting. The November model is described as making that control more precise, especially across longer durations. If you build your reference sets carefully now, they're the inputs you'll test the new model with.

How It Fits Next to Seedance and Kling

Alibaba isn't announcing into a quiet market. ByteDance's Seedance 2.5 already makes 30-second single takes, and Kling's Kling 4.0 landed on each::labs in October with reference-to-video and video editing. Both companies are pushing toward longer, more consistent output. Our Wan 3.0 vs Seedance 2.5 comparison shows how close the current generation already is.

What would set Alibaba's next model apart, if the reports hold, is the narrative claim. Length and consistency are where everyone is heading. Planning a story across shots is a different ambition, and it's the claim to hold every November announcement against, Alibaba's included.

The Questions We'll Answer on Launch Day

When the release note appears, five things will tell you whether it's worth switching. First, the real ceiling: how long a clip it makes, and whether that length holds consistent faces and places to the last second. Second, shots: does it cut on its own, and can you say where the cuts go? Third, inputs: does it keep Wan 3.0's mix of images, video, audio and documents, or ask for something new? Fourth, sound: does dialogue stay in sync across a longer piece? And fifth, access: hosted, open weights, or both.

We'll run our own story tests against those five questions as soon as the model is available, and put the results next to Wan 3.0 so you can see the difference rather than read about it.

Get Ready on each::labs Before It Lands

You don't need the new model to start preparing for it. The fastest teams on launch day will be the ones with a test set ready.

Start with a short story, not a shot. Write a three to five beat scene you actually need: a product reveal with a presenter, a two-character exchange, a before-and-after. Run it today on Wan 3.0 Text to Video as a single 30-second prompt, and again shot by shot. Note where consistency breaks. Those breaks are your benchmark.

Next, build your reference kit. A clean presenter image, a product image, a location image and one short clip of the camera motion you like. Our Wan 3.0 prompting guide shows how to refer to each input by order in the prompt. When Alibaba's next-gen video model arrives, you'll be able to run the same kit through it and see the difference in an afternoon.

Then wire the pipeline once. On each::labs every model sits behind the same API, so a flow you build around Wan 3.0 today should take the new model with a model id change. Sketch it in each::labs flows, and browse the other text to video models to give your test a baseline.

Frequently Asked Questions

When is the Alibaba next-gen video model coming out?

November 2026, according to reports from Alibaba's September 22 cloud conference. As of November 9 there's no official release, and no more precise date has been published.

Is Alibaba's next video model called Wan 3.5 or Wan 4?

Alibaba hasn't named it. The reports only call it a new next-generation video model, and we found nothing official using either name.

What will the new Alibaba video model do better than Wan 3.0?

The reported direction is longer output that stays consistent, precise control of characters, scenes and camera, and director-level thinking that moves from single shots to complete narratives. Specs haven't been published, so treat these as goals until Alibaba releases the model.

Can I use it on each::labs?

Not yet. It's coming to each::labs, and you'll be able to run it through the same API as the rest of the catalog. Wan 3.0 and Wan 3.0 Prime are live on each::labs now and are the best way to prepare.