GPT IMAGE 2.5 | WAN 3.0 | SEEDANCE 2.5 IS NOW LIVE!

GENERATE YOUR VIDEOS WITH UNLIMITED SEEDANCE 2.5 TODAY.

NO EXTRA SUBSCRIPTION

Wan 3.0 guide

Wan 3.0 prompt guide: how to write a prompt that works

A good Wan 3.0 prompt names, in this order, who is in the shot, what they wear and do, where it happens and what the light comes from, one camera move, the sound and the last frame. Wan 3.0 reads up to 20,000 characters and up to 10 reference images, 5 videos and 5 audio clips, so describe what you can see and hear, not how it should feel. One clip runs up to 30 seconds.

By  Founder, Reviral Last updated

Six rules for a Wan 3.0 prompt

  1. 1.Start with the person and the action

    The model weighs the first part of the prompt most. Put who is in the shot and what they do first; camera and light go in the middle, sound at the end.

    “A woman in her early twenties ... walks straight toward the camera across a wet zebra crossing at dusk.”

  2. 2.Name every garment by fabric and colour

    Fabric tells the model how it moves and catches light. Give each piece its material and a colour, down to the shoes.

    “a long tangerine (#F28500) satin trench coat hanging open over a fitted black ribbed turtleneck”

  3. 3.Give every light a source

    Say what the light comes from: a streetlight, a shop window, a skylight. A light with a cause lands where you expect it.

    “amber streetlights from the left, cool blue light from a shop window on the right outlining the coat's edges”

  4. 4.One camera move per clip

    Pick one move (track back, arc, crane) and the lens. Two or three moves in one clip make the picture swim.

    “50mm lens, one slow backward tracking move at her walking pace”

  5. 5.Show it with the body, not with adjectives

    Words like stunning or cinematic do not draw anything. A hand into a pocket, a coat swinging, a gust lifting a collar do.

    “On the third step her left hand slides into the coat pocket”

  6. 6.Write the sound and the last frame

    Wan 3.0 makes sound by default, so say what is heard. End with the frame the clip should stop on.

    “Final image: she stops, the coat swings once and settles against her legs”

A tested prompt and the video it made

This video was made with Wan 3.0 on Reviral from the prompt next to it, exactly as written.

  1. 1. Tangerine satin trench on a wet crosswalk at dusk

    5 s · 720p · 9:16 · 99 credits

    A woman in her early twenties of West African heritage with deep brown skin, a tall rounded black afro and large gold hoop earrings walks straight toward the camera across a wet zebra crossing at dusk. She wears a long tangerine (#F28500) satin trench coat hanging open over a fitted black ribbed turtleneck, wide-leg cream (#F2E8D5) wool trousers and black leather square-toe boots. Each stride throws the open front of the coat back past her knees, and the satin catches the amber streetlight in long highlights that slide down the fabric as it folds. On the third step her left hand slides into the coat pocket; her chin stays level, her eyes stay on the lens, her mouth stays closed. A city bus passes behind her and its headlights sweep across the wet asphalt, lighting the reflection of her boots in a puddle. Camera: 50mm lens, one slow backward tracking move at her walking pace, framed from the knees up and ending on her head and shoulders filling the frame as she stops one step from the lens. Light: amber streetlights from the left, cool blue light from a shop window on the right outlining the coat's edges. Sound: her boot heels on wet asphalt, the bus engine passing, distant traffic. Final image: she stops, the coat swings once and settles against her legs, a gust lifts the collar against her jaw.
    Use this prompt

Questions about Wan 3.0 prompts

How long can a Wan 3.0 prompt be?

Alibaba's API reference allows up to 20,000 characters. Write enough to name the person, the action, the place, the light, the camera, the sound and the last frame.

Should I write Wan 3.0 prompts in English?

English works well and is what the prompt on this page uses. Alibaba's API reference counts each Chinese character or letter as one character.

How do I keep the same person across clips?

Add a photo of them as a reference image (Wan 3.0 takes up to 10) and repeat the same description of their face, hair and clothes in every prompt.

What does one Wan 3.0 clip cost on Reviral?

19.8 credits a second at any resolution: 99 credits for 5 seconds.

More on Wan 3.0

Write your first Wan 3.0 prompt

Open Cinema with Wan 3.0 picked and see the price before you press.