Wan 3.0 AI Video Generator

What Wan 3.0 does well

Image and audio references in one clip

Wan 3.0 accepts up to 10 reference images and 5 audio clips, and Alibaba lets you combine them freely in one generation.

Refer to them in the prompt the way Alibaba does, as Image 1, Image 2 and Audio 1: "the fisherman from Image 1 walks along the pier from Image 2 and speaks the line from Audio 1".

Audio clips have a combined length limit, so trim them to the part you need. Need a reference picture first? Make the character or product on Wan 2.7 Image, then add it as Image 1.

Use material you own or have permission to use, especially faces and voices.

Cinematic 16:9 video keyframe, a triptych of three panels separated by thin dark gutters: left, a plain studio portrait of a bearded fisherman in a dark green knitted cap; middle, a plain photo of an empty wooden pier at dawn with moored boats; right, the same fisherman in the same cap walking along that same pier at dawn as the camera tracks beside him, boats and gulls behind. Consistent face and clothing, soft blue and amber palette, no text, no logos.

Long clips that hold a whole scene

Clips can run up to 30 seconds, long enough for a scene with several actions in a row.

Write the prompt as a short sequence, with each beat in order, and give rough timings when they matter: "first she opens the box, then lifts out the lamp, then switches it on and smiles".

Longer clips cost more, because video is charged per second. Test the idea at a shorter length and a lower resolution before you generate the full take.

Cinematic 16:9 video keyframe, a young man at a sunny wooden desk lifts a brass desk lamp out of an open cardboard box, crumpled paper around it, a potted plant and a notebook nearby, the lamp already glowing warm. Natural afternoon light, handheld but steady camera, cozy apartment setting, no text, no logos.

Sound made with the picture

Wan 3.0 generates sound along with the video, so footsteps, rain, engine noise or a spoken line arrive in the same clip.

Describe the sound next to the action it belongs to: "the kettle clicks off and starts to whistle", "she says quietly, 'We made it'".

When a clip is built on references, an audio clip can set a voice or a rhythm for the scene. Alibaba notes that the first-frame and first-and-last-frame modes do not take audio references, so use references when the voice matters.

Listen to every result before you publish it.

Cinematic 16:9 video keyframe, a street drummer in a denim jacket plays a set of upturned plastic buckets with wooden sticks in a busy stone square, pigeons scattering into the air, passers-by stopping to watch, a fountain behind. Bright afternoon light, energetic low angle, motion in the sticks, no text, no logos.

How to write prompts for Wan 3.0

1. Write the shot in order: subject, action, place, camera move, light, then sound.

2. Refer to references by type and order, as Image 1, Image 2 and Audio 1, and say what each one controls.

3. For long clips, list the actions as a sequence and give rough timings for the moments that matter.

4. Describe each sound in the same sentence as the action that causes it, and put spoken lines in quotation marks.

5. Try the idea on Wan 3.0 Prime at a short length and low resolution, then generate the final take once the prompt is settled.

Videos made with Wan 3.0

• 

  Bike ride along the canal

  Generated with Wan 3.0

  A single continuous shot along a rain-washed canal street at dusk. A woman in a yellow raincoat cycles past glowing shop windows, rings her bell at a passer-by with an umbrella and turns onto a small stone bridge over the water. The camera tracks beside her at shoulder height, then swings around behind her as she crosses the bridge. Reflections ripple in the puddles, light drizzle, warm shop light against a deep blue evening sky, the hiss of tires on wet stone and rain dripping from the awnings, cinematic natural look.

• 

  Noodles across the counter

  Generated with Wan 3.0

  A single continuous shot at a small noodle bar at night. The cook lifts a tangle of noodles from a steaming pot, drains them with two sharp shakes, lays them into a deep bowl, ladles hot broth over them and slides the bowl along the wooden counter to a customer, who leans in over the steam. The camera glides slowly along the counter, following the bowl. Warm lamplight, rising steam, the bubbling pot, the clink of the ladle and quiet chatter in the background, cinematic.

Sample prompts to try with Wan 3.0

• 

  Morning at the bakery

  Cinematic 16:9 video keyframe, a medium shot of a baker in a white apron sliding a wooden peel of golden crusty loaves out of a glowing stone oven, flour drifting through slanted early morning light, shelves of bread behind her, steam rising from the loaves. Warm amber and cream palette, slow push-in, documentary feel, no text, no logos.

• 

  Train across the desert

  Cinematic 16:9 video keyframe, an aerial tracking shot following a long silver passenger train as it crosses a red sand desert at sunset, its long shadow stretching over the dunes, distant mesas on the horizon, a thin trail of dust behind the last car. Deep orange and violet sky, smooth drone motion, epic and calm, no text, no logos.

• 

  Small robot in a workshop

  Cinematic 16:9 video keyframe, a small round toy-like robot with one glowing blue eye and stubby arms reaches for a wrench on a cluttered wooden workbench, jars of screws, a desk lamp and blueprints around it, warm lamp light and cool window light mixing. Stylized 3D animation look, gentle camera dolly to the side, playful mood, no text, no logos.

• 

  Violinist on a rooftop

  Cinematic 16:9 video keyframe, a young woman in a long grey coat plays the violin on a city rooftop at blue hour, the wind moving her hair, string lights on a railing beside her, the skyline lights coming on behind. Slow orbit around her, deep blue and warm gold palette, shallow depth of field, no text, no logos.

Sample images are illustrative; results vary by model.

Wan 3.0 specs and pricing

Every price below comes from the live price list, so it changes here as soon as rates change. Credits are charged per image, or per second of video.

Wan 3.0 limitations

• A first frame or a first and last frame cannot be combined with references in the same clip, and those frame modes do not take audio. Choose one approach per clip.

• A long clip leaves more room for faces, hands and small objects to drift.

  Check every second before you publish, and split a scene into shorter clips when one part keeps going wrong.

• The same prompt gives a different clip each time, and a fixed seed does not guarantee a repeat. Keep the take you like rather than expecting to regenerate it.