Alibaba Wan 3.0 · public beta
Wan 3.0 is Alibaba's omni-modal video model. Feed it a prompt, a photo, a clip, a voice note, a slide deck or a URL, and it returns a single continuous take up to 30 seconds long with audio already mixed in.
Community Wan 3.0 clips, unedited. Hover to play, click for the prompt.
What it does
Thirty seconds changes what you can tell in one shot. Wan 3.0 gives a scene room to breathe, a camera room to travel, a character room to finish a thought.
Generate a clipWan 3.0 renders up to 30 seconds as one continuous take — double the ceiling of the model before it. Long enough for a real scene instead of a loop.
No stitching, no crossfades hiding a seam — Wan 3.0 renders it in one pass.
Text, photos, video, audio, a PDF, a slide deck, a web page. Wan 3.0 reads all of them as source material and builds the video from what it finds.
Point it at a product page and get a promo cut.
Vertical for TikTok and Reels, square for marketplaces, wide for YouTube. Pick the ratio before you render so nothing gets cropped later.
Six ratios, or let Wan 3.0 match your source footage.
The same prompt box covers cel-shaded animation, gritty live action, and skin-level realism. Wan 3.0 follows the style you describe instead of forcing a house look.
Say the look you want and it holds.
Drift is what kills long AI clips — a character morphs, a logo warps, a room rearranges itself. Wan 3.0 was built to hold characters, products and spaces steady across the whole take.
Thirty seconds is only useful if second 29 still matches second 1.
The short version
Wan 3.0 is the video generation model Alibaba released into public beta in August 2026. It produces single-shot clips of up to 30 seconds at 480p, 720p or 1080p, with a synchronized audio track and multilingual speech, from text or from mixed media you supply.
What separates Wan 3.0 from most video models is the range of things it accepts as input. Alongside prompts and images it reads video, audio, PDFs, presentations and live web pages, which makes it useful for turning material you already have into something watchable. On this site Wan 3.0 runs in the browser and bills in credits — a 5-second 480p clip costs 10 credits, and 1080p costs 38. Failed renders are refunded, and there is no API key or subscription to set up.
Picking between them
Both reach 30 seconds. Wan 3.0 differs on what you can put in, and on what you pay.
| Wan 3.0 | Seedance 2.5 | |
|---|---|---|
| Max length | 30 seconds | 30 seconds |
| Resolution | Up to 1080p | Up to 720p |
| Audio | Native, multilingual speech | Native, optional |
| PDF / slides / web page input | Yes | No |
| Reference images | Up to 10 | Up to 30 |
| Cost | From 10 credits per 5s | From 30 credits per 5s |
Rule of thumb: reach for Wan 3.0 when the source is a document, a web page or mixed media, or when 1080p matters. Reach for Seedance 2.5 when you need a large reference stack.
Where it earns its keep
The long single take and the document input make Wan 3.0 worth reaching for on jobs that were not worth starting before.
Hand Wan 3.0 a PDF or a slide deck and get a narrated video back. Useful for course modules, onboarding, and the internal update nobody was going to read.
Thirty unbroken seconds fits a whole beat — setup, turn, punchline. Wan 3.0 holds the actor's face and the set steady long enough for the scene to land.
Getting started
Three steps. No install, no key, no card.
Type a prompt, or upload a photo, a clip, a voice note, a document, or paste a web page URL. Wan 3.0 accepts them mixed together.
Choose anywhere from 2 to 30 seconds, pick a resolution, and pick a ratio. Wan 3.0 updates the credit cost as you change them.
Wan 3.0 returns a finished clip with its audio track. Keep it, tweak the prompt and rerun, or post it as is.
Getting better output
Small changes to how you write the prompt move a Wan 3.0 result more than any setting does.
Thirty seconds is long enough to need structure. Tell Wan 3.0 where the shot starts, what changes in the middle, and where it lands.
Audio is generated with the picture, so describe it. Put spoken lines in quotes, name the music mood, name the room tone.
Lock the composition at 480p for 10 credits, then rerun the prompt you like at 1080p. Paying 38 credits for a draft you throw away is the expensive way.
A 40-page PDF gives Wan 3.0 too much to choose from. Feed it the two pages that matter and the cut gets tighter.
Questions
What people ask before their first render.
Your turn
A prompt, a photo, or the deck sitting in your downloads folder. Wan 3.0 turns it into a clip with sound in a couple of minutes.
