Wan 3.0 is the newest generation of Alibaba's Wan video models. Alibaba announced it in August 2026 and made it available on Alibaba Cloud Model Studio. Its headline features are 30-second single generations, "everything to video" input (including documents) and noticeably more lifelike faces. This guide summarizes what it does, what it costs, and one point many pages get wrong: its license.
Wan 3.0 at a glance
- Max length: 30 seconds per generation, with smart duration recommendation and video extension.
- Inputs: text, images, audio, video and documents (doc, xls, ppt, pdf, txt, key, pages, numbers, md; one file up to 100 MB and 50 pages).
- Resolutions: 480p, 720p and 1080p.
- Editing: change visuals, plot and dialogue of a generated video without starting over.
- API price: $0.05 per second (480p), $0.10 (720p), $0.20 (1080p). A 30-second 1080p clip costs $6.00 at list price.
Source: Alibaba Cloud, Wan 3.0 announcement.
Is Wan 3.0 open source?
No. Wan 3.0 is offered through Alibaba's API. The open-weight Wan models released under Apache 2.0 are earlier versions (Wan 2.1 and Wan 2.2), which you can find on Wan-Video on GitHub and Hugging Face. If you want to run Wan locally, those older versions are your option; for Wan 3.0 you need an API provider.
What's new compared with Wan 2.x
30-second storytelling
Earlier Wan versions produced short clips. Wan 3.0 doubles that to 30 seconds and recommends a duration based on your prompt, so a simple product shot is not stretched and a narrative is not cut short.
Documents as input
Wan 3.0 is the first Wan model that reads documents. Alibaba's examples include turning a product deck into a launch ad, a training deck into courseware, and a spreadsheet into animated charts. Pair the document with a one-sentence instruction such as "Turn this brand story into a warm 30-second brand film."
More realistic people
Alibaba highlights "diverse, lifelike faces" and natural micro-expressions in text-to-video, plus stronger consistency for characters, props, spaces and style in reference-to-video mode.
Known limitations
Alibaba itself notes that audio texture and on-screen text rendering are still improving. If your video depends on readable text, add it in editing instead.
How to prompt Wan 3.0
- State the length and purpose: "30-second product teaser for Instagram, 9:16."
- Give a three-part structure: opening, development, closing beat.
- Say what references are for: "Keep the character's face and outfit from image 1; use image 2 for the room."
- Draft at 480p, finish at 1080p. At list prices a 30-second draft costs $1.50 at 480p versus $6.00 at 1080p.
More templates: 30 text to video prompt ideas.
Wan 3.0 vs other 30-second models
Wan 3.0 is one of three major models that reached 30-second single generations in 2026, alongside Kling 4.0 and Seedance 2.5. Wan's distinguishing feature is document input; Seedance leads on reference volume; Kling 4.0 adds multi-keyframe control. See all options in our Sora alternatives guide.
Using Wan on TextVidAI
TextVidAI lets you run prompts across several video models from one account. Check the TextVidAI generator for the Wan version currently available.
FAQ
How long can Wan 3.0 videos be?
Up to 30 seconds per generation, with extension to continue a story.
How much does Wan 3.0 cost?
On Alibaba Cloud Model Studio: $0.05 per second at 480p, $0.10 at 720p and $0.20 at 1080p. Third-party platforms set their own prices.
Can I download Wan 3.0 weights?
No. Only earlier Wan versions (2.1 and 2.2) have open weights.
Sources
Last updated: October 9, 2026.