WAN 3.0
Sign inStart free
Yearly billing

Yearly is half the monthly rate — the same credits either way

See the plans

Wan 3.0 prompt examples, printed in full

Every one of these Wan 3.0 prompt examples is the exact text that produced the clip beside it — nothing summarised, nothing shortened. Each is broken into the six layers Wan 3.0 reads a prompt along: subject, scene, motion, camera, sound and style. Copy one, change a noun, run it.

11 templates · every prompt complete, beside the clip it produced

Text to video

2

Nothing to upload. The prompt is the only source of truth, so it has to carry the subject, the setting, the motion and the look — which is why these two are the longest text on the page. Run one in Text to Video.

  • A ruined overgrown avenue, a child holding a light in a derelict store, and an astronaut lifting off her helmet
    P—01Text only

    Astronaut in the ruins

    The plainest shape a Wan 3.0 prompt takes: what happens, then what it looks like.

    5s720PWan 3.080 cr
  • A young man and an elderly woman play chess at a table in a busy city square
    P—02Text only

    Street chess turnaround

    A reversal in five seconds, carried by one face rather than by anything on the board.

    5s720PPrime120 cr

Image to video

2

A first frame already holds the subject, the setting and the style, so the prompt only says what changes. These are the two shortest prompts here, and that is the lesson rather than an accident. Run one in Image to Video.

  • A small ranger reaches toward an enormous green dragon lying in a sunlit forest clearing
    P—03Needs an image

    The dragon opens its eyes

    The shortest prompt here. A first frame already holds the world, so the text only says what moves.

    5s720PWan 3.080 cr
  • A man with a camera stands in a busy European square looking toward a distant clock tower
    P—04Needs an image

    The man at the clock tower

    A five-second story hook that never describes its subject, because the first frame already did.

    5s720PPrime120 cr

Reference to video

3

Up to ten images, five video clips and five audio clips in one request — the mode Wan 3.0 was built around, and the only one with syntax the model actually parses. Run one in Reference to Video.

  • A masked rider on a black horse gallops alongside a steam train across a sunset desert
    P—05Needs references

    Desert train heist

    Four action beats written into a five-second request — and the two ways to fix that.

    5s720PWan 3.080 cr
  • A silver-haired navigator holds a cyan star map opposite a black-coated operative in an orbital scrapyard
    P—06Needs references

    Orbital scrapyard handoff

    Paragraph beats and a full maintain clause — with the asset numbering it should have had.

    5s720PPrime120 cr
  • A woman in white sits on a lime-green sofa in a flower meadow as a butterfly lands on her hand
    P—07Needs references

    Meadow sofa portrait

    A near-still portrait that holds a reference image while only the butterflies move.

    5s720PWan 3.080 cr

Past five seconds

4

Twelve and fifteen seconds, where the writing changes. Under eight seconds a prompt is one action; past that it has to name its beats in order and say how it ends, because Wan 3.0 fills every second you pay for and invents an ending for the ones you left blank. Run one in Text to Video.

  • Five dancers in black dresses silhouetted against a hard white backlight in ground fog
    P—08Text only

    Dancers in backlit fog

    One hard backlight, five silhouettes, and three ordered beats: gather, open, close.

    12s720PWan 3.0192 cr
  • A red supercar unfolding into a four-legged machine on wet asphalt under floodlights
    P—09Text only

    Supercar to predator

    A car becomes an animal in four ordered beats, the last of which is the ending.

    15s720PWan 3.0240 cr
  • A pink mech suit firing a beam at a green armoured brute across a concrete plaza
    P—10Text only

    Schoolgirl to mech suit

    Street, then a face behind glass, then a stand-off — three shot sizes in one request.

    15s720PWan 3.0240 cr
  • A first-person view charging across a blue glacier canyon past a wrecked ship, with ice crusting over both arms
    P—11Text only

    Across the glacier

    A continuous first-person travel shot where duration buys distance rather than beats.

    15s720PWan 3.0240 cr

Six of these clips are Alibaba's own published example for that model, so the prompt really did produce the clip — and the run was chosen by the people selling the model. The other five were published elsewhere with no prompt attached, and their prompts were read back out of the footage. Every card says which it is and links to its source.

Now the small print

Free to copy, nothing gated

Every Wan 3.0 prompt above is complete and visible — no truncation, no paraphrase, no "unlock the full prompt", no watermark over the clip. Take them, change the nouns, use them somewhere else if you like. A prompt is plain text and we do not think plain text should have a paywall.

Free here, no account

Costs credits

  • Generating a clip on Wan 3.0 — your first one is on us: 480P, up to three seconds
  • 720P and 1080P, and anything longer than that
  • Reference video, which bills at the same rate as the output

The split is not generosity. Text costs us a fraction of a cent to produce and a clip costs real money from the first second, so the honest thing is to charge for one and not the other.

Four steps

How to use one of these

Copying a whole prompt and changing the subject works, and it is what most people do first. Taking one layer works better, and it works on subjects nothing like the original.

The six layers

  • Entity
  • Scene
  • Motion
  • Aesthetic control
  • Stylization
  • Sound
  1. 01

    Find the motion, ignore the subject

    Pick the clip whose camera and movement are closest to what you want. What is in frame does not matter yet — you are shopping for a shape, not a scene.

  2. 02

    Take two layers verbatim

    Open the breakdown and copy the Aesthetic control and Sound lines exactly as written. Those two are what most people write worst, and they transfer to any subject without editing.

  3. 03

    Write your own Entity and Scene

    Who is in it and where it happens. This is the part only you know, and it is the only part worth spending your own words on.

  4. 04

    Keep the length, or rewrite for the new one

    Wan 3.0 takes any whole number of seconds from 2 to 30, and the prompt shape changes with it. Under 8s write one action. Past 15s name the beats in order and say how it ends — an unwritten ending gets invented.

The four-layer swap is the fastest way to a result that looks directed rather than generated. It is also how the prompt generator on this site works, if you would rather have it done for you.

Write your own instead

If nothing here matches, the Wan 3.0 prompt generator builds one from a single line of description, in the same six-layer structure these examples use. Free, no account, no daily cap.

  1. 01

    Describe it in one line

    A sentence is enough. The generator asks for the length and the mode, not for a paragraph.

  2. 02

    It composes the six layers

    Entity, scene, motion, aesthetic control, stylization, sound — flattened into the plain prose Wan 3.0 reads.

  3. 03

    Send it straight to the generator

    The prompt, the length and the ratio travel together, so nothing has to be re-typed.

Open the prompt generator

Or start from a video you like

Already have a clip whose look you want? Video to prompt reads it and writes the Wan 3.0 prompt backwards from the footage — also free, also no counter, and it works on your own material rather than on ours.

Open video to prompt

Both tools are text and audio models, not video ones. They cost us a fraction of a cent a run, which is why they carry no counter while generating a clip does.

Before you copy anything

Wan 3.0 prompt questions

Ten answers · all visible · nothing collapsed

Are these Wan 3.0 prompts free to copy?

Yes. Every prompt on this page is printed in full and free to take, edit and use. No account, no email, and copying never spends credits. Generating a clip does.

Are the prompts complete, or truncated?

Complete. Each one shows its character count on the same rule as the text, so you can check what you pasted against what we published. That number is computed from the string, not typed in.

Were these clips generated with Wan 3.0, using these prompts?

Six of them, yes — those are Alibaba's own published examples, so the prompt and the clip are a real pair, and also a vendor render chosen by the people selling the model. The other five were published without a prompt, so we read one back out of the footage and the card says so. Neither kind was generated on this site yet.

Does Wan 3.0 need a structured prompt?

No, and this is the mistake that costs people the most. Wan 3.0 reads one plain string of up to 20,000 characters. Field names, [Shot 1] markers and 00:00.000 timestamps are not parsed — they are rendered into the picture, and the request still succeeds.

How do I write dialogue?

Put the exact words in braces — {It is not ready.} — and leave the speaker and the delivery outside them in ordinary prose. Braces are the whole mechanism Wan 3.0 uses to tell a spoken line from a described one; quotation marks get the line narrated instead.

How do I exclude something? There is no negative prompt field.

Correct, there is not. Write the exclusion as a sentence inside the prompt: "no music", "no on-screen text". Anything formatted as a negative prompt arrives as words to render.

Why do some copied prompts come back wrong?

Usually length. Wan 3.0 generates 2 to 30 whole seconds, and a prompt describing five beats in a five-second request is not truncated — the whole schedule is compressed into the runtime you asked for, so every beat arrives early or most of them are dropped. Raise the duration rather than cutting the prompt.

Will the same prompt give me the same clip twice?

Close, not identical. Wan 3.0 takes a seed, so fixing it gets you much nearer — but a different resolution still changes the result. Expect the same register rather than the same frames.

How do I point a prompt at an image I uploaded?

Write Image 1, capitalised with a space, numbered by the file's position in the array. Video 1 and Audio 1 work the same way. Lower case, underscores or "picture 1" bind to nothing, and the model quietly invents the subject instead.

What does running one of these cost?

It is printed on every card, in credits, computed from the seconds, the resolution and the model tier by the same code that debits your balance. The three resolutions double at each step and there is no free floor on length, so a fifteen-second example costs three times its five-second neighbour, and any of them costs half as much at 480P.

Take one and run it.

Copying needs no account at all — nothing here is gated, truncated or watermarked. Generating needs an email, and your first clip is on us: Wan 3.0 itself at 480P, up to three seconds, with sound.

Sign up free · 1 clip on Wan 3.0

Written and maintained by the wan-3.run editorial teamPublished Last updated