Wan 3.0 AI Video Generator

Turn a prompt or your images into a video with sound, up to 30 seconds long. Wan 3.0 and the faster Wan 3.0 Prime by Alibaba, online in your browser, in up to 1080p and without a watermark.

Try Wan AI online

Model

0/20,000

Aspect ratio

Quality

5s

Audio

5s · 480p · with audio · No watermark

Example · Wan 3.0
  • Wan 3.0
  • Text to Video
  • Image to Video
  • Reference to Video

A Wan 3.0 video. Write a prompt or add your images to make your own.

What is Wan 3.0?

Wan 3.0 (also written Wan3.0) is the newest Wan AI video model from Alibaba, the latest step in a model family Alibaba first introduced in July 2023. It makes a video with its own sound from a text prompt, from a first and last frame, or from reference images. Alibaba opened Wan 3.0 as a public beta on August 6, 2026 and released it on August 24, 2026. On Melies AI you can use it as an online Wan AI video generator, straight from your browser: no Alibaba Cloud account, no API key and no GPU of your own.

Alibaba built the release around four things: native 30-second generation, omni reference, real-world fidelity and dynamic motion. One generation runs up to 30 seconds, twice the 15 seconds of Wan 2.7, so there is room for pacing, continuous camera moves and one-take scenes: in Alibaba’s words, “not a clip, but a complete story”. Faces look like distinct people rather than the same AI face, with natural skin and micro-expressions, and characters, props, spaces and style stay close to your references from start to finish.

Two versions sit in the generator’s model menu. Wan 3.0 is the standard model; Wan 3.0 Prime is Alibaba’s speed tier of the same model, for when getting the finished video sooner matters more, at a little more per second. Both make 2–30 s clips in 480p, 720p or 1080p, in 6 aspect ratios, from prompts of up to 20,000 characters and up to 10 reference images, with sound you can switch off. Every video downloads as an MP4 without a watermark.

Clip length
2–30s
Top quality
1080p
Reference images
10
Sound and speech
Native
Aspect ratios
6
Credits · 5 s at 480p
10

Wan 3.0 and Wan 3.0 Prime by Alibaba · run through a third-party API · not affiliated with Alibaba · Last updated

Wan AI video features

What Alibaba launched Wan 3.0 with, and how to use each part here. The generator above is already set to Wan 3.0; switch to Wan 3.0 Prime for faster renders.

PromptOne 2–30 s take with sound
01

Native 30 seconds: a whole story in one take

Wan 3.0 generates up to 30 seconds in a single pass, twice Wan 2.7’s 15 seconds. That leaves room for narrative pacing, continuous camera moves and unbroken one-take shots, so a short scene with several beats comes out complete instead of stitched from clips. Pick any length from 2 to 30 seconds.

PromptDistinct faces with natural expressions
02

Lifelike faces, not the same AI face

Alibaba’s goal for text to video was “no more same-face AI people”. Wan 3.0 renders distinct faces with natural skin and restrained emotion, ties micro-expressions to body language, and gives each person in a group scene their own reaction. Describe who your characters are and how they feel, not only what they do.

@Image1 + @Image2A new scene, both on-model to the end
03

Characters, props and space stay consistent

In reference to video, Wan 3.0 copies the details of your references instead of a rough likeness: a character’s face, hair, build, clothing and accessories, a product’s shape, logo and materials from every angle, and where everyone stands in the scene, without one style bleeding into another. Alibaba’s demo keeps a cat and a wolf from two images on-model through 30 seconds of fast fighting. Add up to 10 images and call them @Image1, @Image2 and so on.

PromptSpeech with matching lip sync
04

Native sound, voices and lip sync

Picture and sound are generated together. Wan 3.0 speaks in natural voices in many languages and keeps a character’s voice and lip sync steady: in Alibaba’s demo one singer switches between eight languages without losing the beat. Put spoken or sung lines in quotes and say the language, or switch sound off for a silent clip at the same price.

PromptThe same model, in the style you named
05

From live action to animation, UI and charts

One model covers photoreal footage, animation and digital content: Alibaba built Wan 3.0 to show software interfaces, charts and motion graphics clearly and stably, for UI demos, feature animations and data stories. Name the look you want, such as live action, 3D animation or a clean product UI.

Wan AI standard vs Prime

The same model, the same modes, lengths and formats. Alibaba offers Prime as the faster service, for when end-to-end speed matters more; the standard Wan 3.0 costs less per second.

ModelWan 3.0Wan 3.0 Prime
Clip length2–30 s2–30 s
Quality480p, 720p, 1080p480p, 720p, 1080p
Aspect ratiosAuto, 16:9, 9:16, 1:1, 4:3, 3:4Auto, 16:9, 9:16, 1:1, 4:3, 3:4
Reference imagesUp to 10Up to 10
ModesText, image and reference to videoText, image and reference to video
SoundOn or offOn or off
SpeedStandardFaster end to end
Price at 480p2 credits / s3 credits / s
Price at 720p4 credits / s6 credits / s
Price at 1080p8 credits / s12 credits / s
Best forClips up to 30 s with sound at a low price per secondThe same clips as Wan 3.0, rendered faster

Comparing Wan 3.0 with Seedance 2.5? See Seedance 2.5

How to use Wan AI

  1. 01

    Choose a mode

    Text to video for a scene from scratch, image to video to start on your picture (and end on another), or reference to video to keep your characters and products consistent. Wan 3.0 is already selected; switch to Wan 3.0 Prime for faster renders.

  2. 02

    Write your prompt

    Write it like a director’s brief: who is in the shot, what happens beat by beat, how the camera moves and what we hear. Put spoken lines in quotes and call your images @Image1, @Image2 and so on.

  3. 03

    Set length, format and quality

    Any length from 2 to 30 seconds, 6 aspect ratios, 480p to 1080p. Draft at 480p, then finish at 1080p. The price is on the button: a 10-second 720p clip is 40 credits on Wan 3.0 or 60 on Prime.

  4. 04

    Generate and download

    Press Generate. The video renders while you keep browsing, then download the MP4, watermark-free.

Wan AI pricing

Pay per second of video, sound included; all three modes cost the same. Wan 3.0 Prime costs a little more for faster renders. Packs from $9.90, no subscription, failed videos refunded.

Quality5 s10 s15 s30 s
Wan 3.0 · Up to 30 s · 1080p
480p · Standard10 credits20 credits30 credits60 credits
720p · HD20 credits40 credits60 credits120 credits
1080p · Full HD40 credits80 credits120 credits240 credits
Wan 3.0 Prime · Faster renders · up to 30 s
480p · Standard15 credits30 credits45 credits90 credits
720p · HD30 credits60 credits90 credits180 credits
1080p · Full HD60 credits120 credits180 credits360 credits

Wan AI FAQ

Something else? Email support@meliesai.com.

Wan 3.0 is the latest generation of Alibaba’s Wan video model family. It makes videos of up to 30 seconds with native sound in one generation, with lifelike faces and characters, props and settings kept consistent with your references. On Melies AI you can use Wan 3.0 and Wan 3.0 Prime in your browser, from text, a first and last frame or reference images, with no API key.

Alibaba opened Wan 3.0 as a public beta on August 6, 2026 and officially released it on August 24, 2026, through Alibaba Cloud Model Studio and partners. It is live on Melies AI now, in the generator’s model menu, along with Wan 3.0 Prime.

Wan 3.0 Prime is the speed tier of the same model: Alibaba recommends it when end-to-end generation speed matters more. Both make 2–30 s clips in up to 1080p from text, frames or up to 10 reference images, with sound. The standard Wan 3.0 costs less (4 credits / s at 720p against 6 credits / s on Prime), so pick Prime when you want results sooner.

Not here. Every Wan 3.0 video uses real GPU time, so generating costs credits. Packs start at $9.90 for 100 credits, enough for up to 50 seconds of 480p video on Wan 3.0. There is no subscription, and if a video fails its credits are returned automatically.

It is priced per second. Wan 3.0: 2 credits at 480p, 4 credits at 720p, 8 credits at 1080p. Wan 3.0 Prime: 3 credits at 480p, 6 credits at 720p, 12 credits at 1080p. A 5-second 480p clip on Wan 3.0 costs 10 credits and a 30-second 1080p clip 240 credits, sound included.

No. Earlier Wan models such as Wan 2.1 were released as open weights, but Wan 3.0 ships with closed weights: Alibaba offers it as a hosted model through Alibaba Cloud and partners, with nothing to download. You cannot run Wan 3.0 locally or in ComfyUI on your own GPU; you use it online, for example on this page.

Both make clips up to 30 seconds with sound from text, frames or reference images. Wan 3.0 costs much less per second (4 credits / s at 720p against 15 credits / s on Seedance 2.5) and starts at 2 seconds. Seedance 2.5 takes up to 30 reference images instead of 10 and has a 21:9 cinematic ratio. Both are in the same model menu, so you can try a prompt on each.

Pick the Reference tab and add up to 10 images, for example a character, an outfit, a product and a location. They are numbered in the order you add them; refer to them in the prompt as @Image1, @Image2 and so on, such as “@Image1 holds @Image2 and sits on the chair from @Image3”.

Yes. Wan AI generates picture and sound together, including natural voices in many languages with matching lip sync. Put a spoken line in quotes to have a character say it. You can switch sound off for a silent clip at the same price, on both the standard model and Prime.

Alibaba’s Wan 3.0 can extend a generation, edit the visuals, plot and dialogue of a video, suggest a length from your prompt, and work from reference videos, audio, documents and web pages. Those features are not available on Melies AI: here Wan 3.0 makes new videos from text, a first and last frame, or reference images, at the length you pick.

Alibaba says two areas are still improving: the texture of the generated audio and the accuracy of text drawn in the picture. Keep on-screen text short, check spelling in the result, and add exact titles or captions in your editor when they must be right.

Up to 30 seconds in one generation, from 2 seconds, on both the standard model and Prime. Pick the length in whole seconds.

Auto, 16:9, 9:16, 1:1, 4:3, 3:4. Auto follows your first frame in image to video and lets the model pick the shape otherwise; choose 9:16 for vertical video.

Treat the prompt as a director’s brief. Say who is in the shot and how they feel, what happens beat by beat across the clip, how the camera moves, the look (live action, animation or UI) and what we hear, with dialogue in quotes. For a 20- to 30-second clip, give it several story beats, not one long beauty shot. Prompts can run up to 20,000 characters.

No, and there is no uncensored mode. Prompts and images go through a content check before Wan 3.0 runs. Adult content and real people without their consent are not allowed; if a video is blocked, its credits are returned.

No. Every Wan 3.0 video uses GPU time, so you pay per second with credits instead of a monthly plan. There is no daily cap: make as many videos as your credits cover. Purchased credits never expire, and you only pay for videos that finish.

Open the generator at the top of this page, where Wan 3.0 is already selected. Pick a mode, write a prompt or upload images, choose length, aspect ratio and quality, and press Generate. The finished MP4 appears under “My videos”.

Yes. Videos you generate are yours to use, including in ads, client work and monetized posts, as long as your prompt and images do not infringe anyone else’s rights and follow our content standards. There is no watermark.

No. Melies AI is a web app: you use Wan 3.0 in the browser and need no API key or code. Developers who want to call the Wan 3.0 API from their own software can go to Alibaba Cloud Model Studio or an API provider.

No. Melies AI is an independent platform that gives you access to Wan 3.0 through a third-party API provider. It is not affiliated with, endorsed by or sponsored by Alibaba. Finished videos stay under “My videos” for 14 days.

Make your first Wan AI video

Write a prompt or upload an image, set the length up to 30 seconds and get a video with sound.

Start creating