Wan 3.0
30-second storytelling, omni-reference creation, and reality-grade rendering.
Loading inspiration gallery.
Wan 3.0 API access is rolling out; generation on this page currently uses Wan 2.7.
What Is Wan 3.0?
Wan 3.0 is Alibaba Wan's latest video-generation model, now in public beta. Its official launch materials center on five upgrades: omni-reference creation, native 30-second generation, pixel-level consistency, a more immersive audiovisual experience, and more precise video editing.
It accepts text, images, audio, and video, and can also read structured sources such as documents, spreadsheets, slide decks, Markdown files, and webpages. Full production API access is still rolling out, so the capabilities below describe Alibaba's public-beta release rather than a Wan 3.0 runtime already connected to Nuzza.
Official Wan 3.0 Features
Create from Media, Documents, and Webpages
Wan 3.0 understands text, images, audio, video, and structured information. Officially listed inputs include doc, xls, ppt, pdf, txt, key, pages, numbers, md, and a URL, enabling product demos, lessons, animated charts, and business presentations from existing material.
Coordinate Rich Reference Material
Alibaba says Wan 3.0 can use up to 20 reference assets. A document or link can be up to 100 MB and 50 pages, with no more than one file or link in that document-input slot. Prompts direct how the supplied people, objects, sound, video, data, and source material should appear.
Generate a Complete 30-Second Sequence
Wan 3.0 can generate a single video up to 30 seconds, leaving more room for narrative rhythm, continuous camera movement, and one-take staging. Its intelligent-duration feature can recommend a suitable length from the prompt, while video extension can continue the story.
Render More Expressive, Believable Scenes
Alibaba describes more varied faces, finer facial and skin detail, restrained emotion, coordinated micro-expressions and body movement, and clearer group emotion. Digital information also receives more deliberate color, chart, data, and interface presentation.
Keep References Stable Across Production Shots
Wan 3.0 is designed to reproduce reference details across characters, props, voices, spatial relationships, and visual styles. Official materials call out stable facial features, clothing, accessories, object structure, logos, materials, positions, camera viewpoints, and cinematic treatment.
Refine Picture, Story, Dialogue, and Sound
Wan 3.0 improves visual realism, image quality, and sound, and supports instruction- or reference-led edits to the picture, plot, and dialogue. Alibaba also states that sound quality and text accuracy still have room to improve, so generated audio and on-screen text require review.
Wan 3.0 Public Beta and API Access
Use an Official Public-Beta Entry Point
Alibaba lists Model Studio and Qwen Cloud among the official public-beta entry points. Wan.video access is also rolling out for members, alongside the Wanxiang web experience and other Alibaba creation products announced for the beta.
Prepare One Prompt and Its Approved Sources
Combine the prompt with the media references the scene needs. For structured material, use one supported file or URL, keep it within 100 MB and 50 pages, and make each requested visual, fact, role, and sequence explicit.
Confirm Region and Pricing Before an API Call
Alibaba's international announcement lists $0.05 per generated second at 480p, $0.10 at 720p, and $0.20 at 1080p. Full API access is rolling out, so verify the current regional contract and billable rate in Model Studio before production use.
Official Wan 3.0 Use Cases
Build Short Drama, MV, Dance, and Vlog Sequences
Alibaba positions the 30-second window, realistic scene reconstruction, and reference consistency for AI film, short drama and comics, music or dance videos, and lifestyle vlogs, where one connected sequence can carry more of the story.
Turn Product Material into Brand Stories
Wan 3.0 is positioned for product presentation and brand storytelling across appliances, cosmetics, automotive, consumer goods, electronics, fashion, and software. Omni-reference inputs let approved product media and source documents guide the result.
Animate Interfaces, Features, Charts, and Data
The model can turn product UI demonstrations, software feature flows, data visualizations, decks, and structured business information into video while preserving the intended visual treatment and making the information easier to present dynamically.
Present Places, Culture, and Local Food
Alibaba highlights realistic scene reconstruction for destination films and cultural communication, including natural scenery, human landmarks, and local cuisine, reducing the amount of large-scale location shooting needed for an early concept or campaign asset.
Use Public-Beta Claims Precisely
Describe Wan 3.0 as public beta with API access rolling out
Do not imply that Nuzza is already generating with Wan 3.0
Review generated sound, on-screen text, facts, data, and logos
Do not treat announced consistency as an accuracy guarantee
Wan 3.0 vs Wan 2.7 vs Google Veo 3.1
Nuzza status
Shown in Models; Wan 3.0 generation is not connected while full API access rolls out.
Available now in Nuzza.
Available now through Nuzza's connected Veo 3.1 workflow.
Duration
Up to 30 seconds in one generation, with intelligent duration and video extension.
2–15 seconds for text- and image-led generation; other operation limits depend on source media.
Current selectable durations are 4, 6, and 8 seconds.
Inputs and references
Up to 20 assets across text, image, audio, video, documents, structured files, and webpages.
Prompt, start frame, first and last frames, reference media, source video, and optional audio in supported modes.
Text, one start image, or matched first and last frames; no dedicated multi-reference flow is exposed.
Audio and output
Reality-grade picture and audiovisual improvements; announced 480p, 720p, and 1080p pricing tiers.
720p or 1080p, with background-music or supplied-audio controls in supported workflows.
Generated-audio control with 720p, 1080p, and 4K choices in the current mapping.
Editing
Instruction- and reference-led changes to picture, plot, and dialogue.
A connected video-edit workflow accepts a source video, prompt, and optional reference image.
Focused on generation and endpoint-frame direction rather than a dedicated source-video editor.
Best fit
Longer, reference-heavy or document-led video workflows once its production API is connected.
A real Wan workflow today for generation, frame control, references, continuation, or editing.
Short-form Google audiovisual generation, first/last-frame direction, or a current 4K option.
Wan 3.0 has the broadest announced reference and document workflow and the longest single-generation duration here, but it is not yet a Nuzza runtime. Use Wan 2.7 for the connected Wan option today, or Veo 3.1 for a connected short-form Google workflow with generated audio and a 4K choice.
Wan 3.0 FAQs
Is Wan 3.0 available in Nuzza now?
Wan 3.0 appears in the Models directory and has this official preview page, but generation is not connected yet. The workspace on this page uses Wan 2.7.
How long can Wan 3.0 videos be?
Alibaba says Wan 3.0 can generate one video up to 30 seconds. It also supports intelligent duration and video extension.
What can Wan 3.0 use as input?
Official materials list text, images, audio, video, doc, xls, ppt, pdf, txt, key, pages, numbers, md, and a URL. The structured-document slot accepts no more than one file or link, up to 100 MB and 50 pages.
How many reference assets can Wan 3.0 use?
Alibaba's public-beta summary says Wan 3.0 supports up to 20 reference assets.
What is the announced international API price?
Alibaba's international public-beta announcement lists $0.05 per second at 480p, $0.10 at 720p, and $0.20 at 1080p. Check Model Studio for the active regional contract before calling the API.
Can Wan 3.0 keep characters and products consistent?
Alibaba says Wan 3.0 improves consistency for character details, props, voices, spatial relationships, and visual style. Outputs still need production review.
Can Wan 3.0 edit an existing video?
The official announcement says it can modify the picture, plot, and dialogue through instructions or references. Nuzza has not yet registered the production endpoint controls.
What limitations does Alibaba mention?
Alibaba specifically says sound quality and text accuracy still have room to improve. Generated audio, captions, interface text, numbers, and claims should be checked before publication.