Alibaba's Wan 3.0: 30-Second AI Video From Text, Images and PDFs

Alibaba Opens the Wan 3.0 Beta: 30-Second AI Video From Text, PDFs and PowerPoint

Alibaba has switched its new video model, Wan 3.0, into open beta on the wan.video platform — the biggest generational jump for the Wan family so far, and a direct challenge to Google's Veo and OpenAI's Sora in the text-to-video race. The launch is verifiable from the primary source itself: the model appears inside the official app under the name wan3.0, complete with an official launch notice showcasing its five headline capabilities and a demo clip hosted on Alibaba's own content delivery network.

The number that tells the whole story is thirty seconds. That is the native clip length Wan 3.0 generates in a single pass — double what Wan 2.5 delivered, according to specialist coverage that accompanied the rollout. For anyone producing short-form content on TikTok, Reels or YouTube Shorts, this is not a spec-sheet curiosity. It is the difference between a clip that dies mid-idea and one that lands a complete scene.

What Is Actually New: The Five Official Capabilities

The official launch notice inside wan.video sums up the new generation in five capabilities. Here they are as documented, then translated into practical workflow language:

• Omni Reference: feed the model a mix of inputs in a single prompt — images, video clips and audio together. Coverage citing the company points to a striking ceiling: up to 10 images, 5 videos and 5 audio clips per request.
• Native 30-Second Duration: long clips are no longer stitched from short fragments. The duration is built into the model's architecture itself.
• Pixel-Perfect Consistency: Alibaba's pitch centers on characters, props and spatial layouts holding their appearance across shots — precisely the weak spot of every current video model, where faces drift and objects morph from one cut to the next.
• Immersive Audiovisual Experience: the official notice points to sound as part of the generation experience, with audio inputs accepted inside the Omni Reference pipeline.
• Precision Video Editing: in-generation editing tools, so a small fix no longer means exporting to a separate editing suite.

Specialist coverage adds two practical touches: the model proposes the video length it thinks fits your prompt, and an official extension tool stretches existing clips to longer runtimes. The input list is equally unusual for this market: PDFs, web pages and PowerPoint decks can all be digested and turned into moving footage. Text and images are no longer the only doors in.

Three Official Ways In

The beta is reachable through three official gates: the wan.video site directly (the easiest entry point for most creators), Alibaba Cloud Model Studio for teams already inside Alibaba's cloud ecosystem, and the Qwen Cloud API for developers wiring generation into their own products. Reports describe two usage tiers — Standard and a faster Prime — with a launch discount on Standard. We are deliberately not quoting specific prices: nothing has been published on a stable official pricing page at the time of writing, and beta-period pricing is a moving target.

Here is the official demo clip the company itself hosts for Wan 3.0 outputs:

The official Wan 3.0 demo clip is hosted directly on Alibaba's own servers: watch it from the official source (mp4 on cdn.wanxai.com).

What This Means for Your Content Workflow

• Short drama and full-length Reels get cheaper: Alibaba explicitly lists "short-form dramas" and "social clips made by creators" among its target audiences, alongside marketing and training video for companies. One complete 30-second scene covers an entire Reel without cuts or artificial speed-ups.
• File-to-video is a real category now: if you prepare a PowerPoint deck or a PDF for a client, turning it into a promotional video becomes a single-prompt job instead of a toolchain — script, scene generation, edit.
• Consistency means repeatable visual identity: the biggest pain in current AI video is a series protagonist whose face changes every clip. If Wan 3.0's consistency claims hold up in practice, you can build one stable character across an entire campaign — a genuine shift in the economics of branded content.
• Audio inside the pipeline: feeding a voice track as reference opens the door to in-tool synchronization instead of exporting everything to a separate dubbing step.

If video generation is part of your daily production line, the smarter workflow is drafting structured scripts first and then feeding them to the model — which is exactly what a tool like ARWriter's research and article writer does, turning one idea into a generation-ready script. For background on Alibaba's model family, see our coverage of the open-weights Qwen 3.8 release and of LTX-2.5, the open video model playing in the same free-tools lane.

Quick Comparison: Wan 3.0 vs What You Probably Use Today

Read this as a snapshot of positioning, not a benchmark battle. Independent, apples-to-apples numbers for video models remain rare and volatile, and dismissing major players on unverified figures would be malpractice.

Honest Limits Before You Dream

• Claims are not evidence: "pixel-perfect consistency" is Alibaba's marketing until long clips prove otherwise. Visual drift remains the industry's failure mode number one, and every vendor stumbles on it.
• It is a beta: expect queues, rate limits and features that shift week to week. Do not schedule a client campaign around a beta tool.
• Pricing is unsettled: launch discounts will settle into different numbers. Budget on the stable, published prices when they arrive — not on the promo.
• Commercial terms: read the platform's usage terms before generating paid client ads; betas often restrict or condition commercial use.
• Non-English prompts: there is no official statement on Arabic or other language prompts yet. English prompts remain the safest route for precise visual direction in this class of tools.

Frequently Asked Questions About Wan 3.0

Is Wan 3.0 free?

An open beta is live on wan.video with two usage tiers (Standard and Prime) according to launch coverage. Alibaba has not published final pricing on a stable official page yet, so check the plans page inside the platform before committing budget.

How do I access Wan 3.0 from outside China?

Go directly to wan.video and select the wan3.0 model in the model list, or use Alibaba Cloud Model Studio with an international account. Developers can use the Qwen Cloud API.

Can I write prompts in Arabic?

There is no official confirmation of Arabic prompt support. In practice, precise visual descriptions in English still outperform other languages in this class of video models — draft in Arabic, then translate the prompt before generating.

What is the maximum clip length?

Thirty seconds in a single native generation per the official launch notice, plus an official extension tool for pushing existing clips longer.

How is it different from Wan 2.5?

Native duration doubled to 30 seconds, inputs expanded to PDFs, web pages, PowerPoint and mixed media references (images, video, audio) in one prompt, with announced improvements in visual consistency and precision editing.

Official Sources

• Official Wan AI platform (wan3.0 model and launch notice): wan.video
• Official demo clip on Alibaba's CDN: cdn.wanxai.com

Bottom Line: Test It Now, Migrate Later

Wan 3.0 is offering something few in this price class do: a long native duration, unusual document inputs and in-tool editing, all verifiable in the official product today. The practical rule for the next two weeks: book two test sessions on the content types you actually sell — Reels, a short ad, a drama cold-open — measure the share of clips that are genuinely publishable, then decide. Either way, the durable skill is writing well-structured long-form video prompts, because that skill survives whichever model wins the race. Start by shaping your idea into a clean script with ARWriter, then feed it to the video pipeline of your choice.