Wan 3.0 on ToVideo
Move past short demo clips. Wan 3.0 targets up to 30 seconds in one generation, native audiovisual output, and multimodal references — including documents — so a single brief can become a complete short story.
Up to ~30s · Native audio · Text · Image · Omni-reference
What is Wan 3.0?
Wan 3.0 (often written Wan3.0) is the August 2026 generation of Alibaba’s Wan video model family from Tongyi Lab, available through Alibaba Cloud Model Studio and related surfaces. Official materials position it as an all-in-one audiovisual model: generate up to about 30 seconds in a single pass (versus shorter caps on earlier Wan 2.x lines), output synchronized dialogue, BGM, and effects, and accept multimodal references — text, images, video, audio, and for the first time in the family, documents and web pages. Resolutions typically include 480p, 720p, and 1080p, with per-second API pricing that rises with quality. Supporting features include smart duration recommendation (auto length when duration is set to adapt), video extension, first-frame / first-and-last-frame control, and carry-over editing from Wan2.7 so you can revise visuals, plot, and dialogue without a full restart. On ToVideo, Wan 3.0 sits in the model catalog as a long-form cinematic option for creators evaluating Alibaba’s latest Wan stack beside other frontier generators — open the AI Video Generator to iterate with credit-aware workflows.
Up to 30 seconds in one continuous generation
Most mainstream AI video tools still think in short hooks. Wan 3.0’s headline upgrade is narrative room: continuous camera moves, multi-beat arcs, and one-take pacing without stitching unrelated five-second fragments. Smart duration helps a simple product beat avoid forced padding, while extension continues a storyline when the first pass ends mid-thought. Plan prompts as full shots — subject, conflict, camera, and sound across time — not a single frozen pose.
Everything to video, including documents
Beyond text and stills, Wan 3.0 reads structured files — doc, xls, ppt, pdf, txt, key, pages, numbers, md — typically one file or link per request (commonly ≤100MB and ≤50 pages). A product deck becomes a launch film; a training PPT becomes courseware; a report becomes a narrated briefing. Pair the document with a clear creative brief so the model knows whether you want a warm brand TVC, a data explainer, or a hard-sell retail spot.
Real-world fidelity and reference consistency
Alibaba materials stress diverse lifelike faces, natural micro-expressions, and stable identity for characters, props, spaces, and style in reference-to-video. That matters when logos, wardrobe, and blocking must survive across a longer take. Bind references explicitly in the prompt (Image 1, Video 1, Audio 1 style naming) so the model knows which asset drives face, motion, or voice.
Native audio plus edit and extend loops
Wan 3.0 natively outputs dialogue, BGM, and effects with the picture. Editing intents (convert, remove, replace, change) and extension intents (extend, continue) let you revise or grow a clip instead of regenerating from zero. Official notes still call out active improvement areas such as audio texture and on-screen text accuracy — draft important supers carefully and verify finals.
Why teams evaluate Wan 3.0
Choose Wan 3.0 when length, multimodal briefs, and production-style consistency matter more than a three-second novelty loop.
Wan 3.0 capabilities that matter in practice
Map Alibaba’s GA strengths to the controls creators actually plan around: length, inputs, audio, frames, and finishing quality.
Text-to-video with cinematic length
Prompt-only generation for blank-slate scenes, with room for multi-beat pacing up to about 30 seconds and native sound.
Image-to-video with first / last frames
Open on an approved still; optionally land on an end frame so product and talent identity stay anchored.
Omni multimodal references
Combine images, short videos, and audio — plus documents or web links where the platform exposes them — for consistency-heavy briefs.
Native dialogue, BGM, and effects
Audio arrives with the picture. Prompt the sonic world explicitly; keep spoken lines short and stageable.
Smart duration and extension
Let the model recommend length when useful, then extend forward or backward to grow the storyline without a cold restart.
480p / 720p / 1080p finishing
Explore at 480p, review at 720p, and finish heroes at 1080p. Cost scales with seconds and resolution.
Wan 3.0 FAQ
Straight answers for marketers and creators evaluating Alibaba Tongyi Wan 3.0.
Explore cinematic AI video on ToVideo
Open the AI Video Generator, select Wan 3.0, choose text, first/last-frame, or reference mode, set duration and quality, and generate with credits visible upfront.