A new inflection point for AI video might be on the way.

On August 6, Alibaba's Tongyi Wanxiang 3.0 (Wan 3.0) officially entered public beta. This update has two key changes: first, it supports generating video directly from office documents like PDF, PPT, Excel, and Markdown; second, single-generation duration has been raised to 30 seconds.

Sounds like just a few more input formats and longer duration — what's the big deal? If that's what you're thinking, you might be underestimating the significance.

Before this, the dominant paradigm of AI video generation was "text description → generate video" — you describe the scene you want in words, and the model generates a few seconds of video. This mode is more like a creative tool, a toy — good for ad assets, short video clips, artistic creation, but hard to use directly for serious productivity scenarios.

Document-to-video changes everything. It means AI video has, for the first time, truly plugged into enterprise knowledge production workflows. Got a product PPT? Upload it and auto-generate a product demo video. Got a data analysis Excel? Upload it and auto-generate a data visualization explainer video. Got a technical Markdown document? Upload it and auto-generate a tutorial video.

This isn't an upgrade of a creative tool — this is a transformation of enterprise content production methods.

From "Second-Long Clips" to "30-Second Narrative"

The 30-second number is also worth discussing on its own.

Previously, AI video generation durations were generally 5–10 seconds — basically just one shot, one scene, one action. Want to tell a complete story? Want a product intro with a beginning, middle, and end? Impossible — you'd have to generate multiple clips and edit them together yourself.

What does 30 seconds mean? It means AI video has narrative capability for the first time. 30 seconds is enough to tell a complete short story, enough to showcase a product's core features and highlights, enough for a short video with a beginning and an end. For social media, 30 seconds is also a sweet spot — not too short to convey information, not too long for users to scroll past.

Of course, compared to professional video teams producing minutes or even tens of minutes of finished content, 30 seconds is still very short. But this is a zero-to-one breakthrough — from "only generating clips" to "generating complete narrative units" — that's a qualitative change, not a quantitative one.

Combined with the document input feature, Wan 3.0's positioning is crystal clear: it's not for making cinematic VFX — it's for solving content production efficiency problems for businesses and individuals. Marketing, education and training, internal onboarding, product presentations, data analysis visualization — these scenarios don't need cinematic picture quality, but they need massive, fast, low-cost video output. And this is precisely where AI video is most likely to land first.

AI Video Competition Enters the Second Half

The competitive landscape of the AI video track is quietly shifting.

The first half was about who could generate more realistic footage, who had better consistency, who could produce more flashy effects. That was the era of Sora, Kling, Runway — everyone was flexing their muscles, competing on technical upper limits.

What's the second half about? Scenario deployment capability. Whoever can find real paying use cases, whoever can integrate AI video into real workflows, whoever can genuinely help users improve efficiency and reduce costs — those are the ones that survive.

Alibaba's entry point is smart: starting from enterprise office documents, directly tapping into the largest existing content library. How many companies in China use PPT for presentations? How many data analysts use Excel for reports? How many tech teams use Markdown for documentation? These are all ready-made content and ready-made demand — turning all of that into video with one click has direct, quantifiable value.

Of course, that's not to say picture quality doesn't matter anymore. Picture quality is foundational — without good quality, everything else is moot. But when everyone's quality reaches a usable baseline, what determines victory is no longer technical parameters — it's scenarios and ecosystem.

AI video is shifting from the question of "does it look good" to "is it useful." And the latter is what truly determines whether this industry can scale.

See you tomorrow.