Playcut v2 is coming MCP · Actor v2 · & much more
Prompt library

Wan 2.7 prompts, written short because Wan expands them

Wan 2.7 prompts for talking photos driven by your audio, scenes, multi-shot clips, reference casts and edits, kept short because Alibaba expands every prompt.

Updated · Playcut prompt library

Keep a Wan 2.7 prompt short and literal, with what must appear first, because Alibaba rewrites and expands every prompt with a language model before Wan renders it. All 28 cards here are written that way, for talking photos driven by your own audio, text-to-video scenes, timed multi-shot clips, reference casts and edits of a clip you already have.

Each chip gives the clip length and its cost in credits on Playcut, at 720p unless it says 1080p.

Wan 2.7 prompts: short, literal, key detail first

Start with who is in the shot, where they are and what moves. Alibaba's text-to-video and image-to-video prompt guide calls that basic formula Entity + Scene + Motion. Its advanced version adds aesthetic control (light, shot size, camera angle, lens and camera movement) and then a style. Each mode has its own version:

Mode What the prompt holds Cards
Text to video Entity + Scene + Motion, then aesthetic control and stylization Text-to-video cards
Image to video Motion + Camera movement (the image already sets the rest) Talking-photo cards
Multi-shot Overall description + Shot number + Timestamp + Shot content Multi-shot cards
Reference to video Reference identifier + Action + Scene Reference cast cards
Video edit The change, then what stays the same Edit cards

The edit row is our advice, not Alibaba's. The guide also warns against real people's names, rapid scene changes in one clip, readable on-screen text and lip-synced prompt dialogue.

Your prompt is expanded before it renders

Every Wan 2.7 job reaches Alibaba with its prompt extension switched on. Alibaba's Wan 2.7 text-to-video API reference describes it: a language model rewrites the input to improve generation quality. The video you get follows that rewritten text, not your exact words.

So lead with the subject, the action and the one detail that matters, keep the wording literal, and on edits name what must not change.

The rewrite is Alibaba's, not a Playcut feature. For the other per-model rules that change a prompt, see our guide to fixing weak prompts.

Talking photos: a still plus your own audio

Run on Wan 2.7

Attach a still plus a WAV or MP3 of 2 to 30 seconds (15 MB at most), and write only the motion and the camera. The audio carries the words. Audio shorter than the clip leaves the rest silent, so match the lengths.

  1. Museum guide

    10 s · 400 cr
    The museum guide speaks to the camera with a calm, friendly expression, small nods on key words, one hand gesturing toward the painting behind her. Steady medium shot with a very slow push-in.

  2. Painted ship's captain

    8 s · 320 cr
    The captain in the oil painting speaks with a gruff, amused look, eyebrows lifting and beard moving with the words, the brushwork staying visible. The camera stays still, framed like the canvas.

  3. Podcast host

    15 s · 600 cr
    The host leans slightly toward the microphone and talks with relaxed energy, glancing down at notes now and then, shoulders loose. Locked-off medium close-up, background softly out of focus.

  4. Teacher at the whiteboard

    12 s · 480 cr
    The teacher explains to the class, turns briefly to point at the diagram on the whiteboard, then looks back to camera with an encouraging smile. The camera pans a little to follow the gesture.

  5. Plush toy narrator

    10 s · 400 cr
    The plush rabbit tells a bedtime story, its stitched mouth moving softly, ears twitching now and then, a gentle sway as if held on a lap. Slow push-in, lamp light as in the photo.

  6. Founder update

    8 s · 1080p · 480 cr
    The person in the headshot gives a short update to camera with natural blinks, light head movement and a confident smile at the end. The framing stays fixed.

Text to video scenes

Run on Wan 2.7

Each card runs who, where and what moves, then light, shot, lens and camera, then style. Text to video takes 16:9, 9:16, 1:1, 4:3 or 3:4 and 2 to 15 seconds.

  1. Stonemason at work

    5 s · 200 cr
    A stonemason in a denim apron chisels a curling leaf into a block of pale limestone in a sunlit yard, dust drifting through the light. Close-up, low side light, shallow depth of field, slow push-in, warm documentary look.

  2. Night bus in the rain

    8 s · 320 cr
    A night bus pulls into an empty country stop in heavy rain, its headlights cutting through the drops, and one passenger steps down under an umbrella. Wide shot, low angle, sodium street light, fixed camera, moody cinematic grade.

  3. Florist wraps a bouquet

    9:16 · 6 s · 240 cr
    A florist ties a bouquet of orange ranunculus on a zinc counter, then slides it toward the camera. Medium shot, soft overcast window light, 50 mm lens, gentle handheld, clean bright look.

  4. Foxes at dawn

    10 s · 1080p · 600 cr
    A family of foxes slips out of a hedgerow at dawn and crosses a frosty field in single file. Aerial shot, slow forward glide, low golden sun, long shadows, nature-documentary style.

  5. Origami crane takes off

    1:1 · 5 s · 200 cr
    An origami crane unfolds its wings on a wooden desk and flaps up into the air, the paper creases catching the light. Macro close-up, top light, dark background, slow orbit, stop-motion feel.

  6. Bucket drummer at rush hour

    8 s · 320 cr
    A street drummer plays upturned buckets outside a train station at rush hour while commuters stream past. Medium wide shot, late-afternoon light, handheld tracking to the left, energetic urban look.

Multi-shot clips with timestamps

Run on Wan 2.7

Open with one line about the whole clip, then number each shot and give it a time range. Give each shot about three seconds or more, as Alibaba's own example does, and name every hard cut.

  1. Chess in the park

    10 s · 400 cr
    A chess game in the park in three shots. Shot 1 [0-3 s] Wide: two old men at a stone chess table under plane trees. Shot 2 [3-7 s] Hard cut, close-up: a hand hovers over a knight, then moves it. Shot 3 [7-10 s] Hard cut, medium: the winner laughs and resets the pieces.

  2. Commute by bike

    12 s · 480 cr
    A commuter's ride to work in three shots. Shot 1 [0-4 s] Aerial wide: a cyclist crosses a river bridge at sunrise. Shot 2 [4-8 s] Hard cut, low tracking shot beside the front wheel on wet cobbles. Shot 3 [8-12 s] Hard cut, medium: she locks the bike outside a glass office and walks in.

  3. Storm over the vineyard

    12 s · 1080p · 720 cr
    A summer storm reaching a hillside vineyard in three shots. Shot 1 [0-4 s] Wide: dark clouds roll over rows of vines, wind turning the leaves silver. Shot 2 [4-8 s] Hard cut, close-up: the first heavy drops hit a crate of grapes. Shot 3 [8-12 s] Hard cut, wide: the winery's lit windows glow through sheets of rain.

  4. The lost mitten

    14 s · 560 cr
    A small lost-and-found story in three shots. Shot 1 [0-4 s] Medium: a girl searches under park benches at dusk, worried. Shot 2 [4-9 s] Hard cut, close-up: a scruffy dog trots up with her red mitten in its mouth. Shot 3 [9-14 s] Hard cut, wide: she kneels and hugs the dog as the street lamps come on.

Reference casts with Image 1 (no brackets)

Run on Wan 2.7

Name the images Image 1, Image 2 and so on, in attach order: capital I, a space, no brackets. Five at most. Repeat each character's own label instead of he or she.

  1. One character, one place

    2 refs · 8 s · 320 cr
    The beekeeper in the white veiled hat from Image 1 walks between the hives in the orchard from Image 2, lifts a frame and holds it up to the light as bees drift around. Medium tracking shot from the side, soft morning light.

  2. Product in a setting

    2 refs · 6 s · 240 cr
    The ceramic teapot from Image 1 sits on the low tatami table from Image 2; a hand pours tea into a small cup and steam curls up. Close-up, soft side light, slow push-in.

  3. Two characters and a porch

    3 refs · 10 s · 400 cr
    The postman in the red cap from Image 1 hands a small parcel to the girl in the yellow dress from Image 2 on the porch of the house from Image 3. The girl in the yellow dress tears it open and laughs. Medium two-shot, afternoon sun.

  4. Mascot in a shop

    2 refs · 8 s · 320 cr
    The felt dinosaur toy from Image 1 waves from the counter of the bookshop from Image 2, then hops down and shuffles past the shelves. Low-angle tracking shot, warm lamp light.

  5. Two riders, two places

    4 refs · 10 s · 400 cr
    The cyclist in the orange jersey from Image 1 and the cyclist in the blue jersey from Image 2 climb the mountain road from Image 3 side by side. The cyclist in the blue jersey pulls ahead at the hairpin by the chapel from Image 4. Aerial tracking shot.

Edit instructions for a clip you already have

Run on Wan 2.7

Upload a 2 to 10 second MP4 or MOV (100 MB at most), then say what changes and what stays. The output runs 2 to 10 seconds in the source's frame. Chips count source plus output seconds, both billed: a 6-second source edited into 6 seconds at 720p is 480 credits.

  1. Day to dusk

    6 + 6 s · 480 cr
    Change the time of day to dusk: a warm low sun, long shadows and street lamps flickering on. Keep the people, their movements and the camera path unchanged.

  2. Recolour a jacket

    8 + 8 s · 640 cr
    Make the rider's black jacket bright red. Keep the fabric folds, the riding motion, the background and the lighting unchanged.

  3. Snowfield background

    10 + 10 s · 800 cr
    Replace the background behind the dancer with a wide snowfield under a pale sky, light snow falling. Keep the dancer, the choreography and the framing unchanged.

  4. Watercolour restyle

    5 + 5 s · 400 cr
    Restyle the whole clip as a loose watercolour painting with soft, bleeding edges and paper texture. Keep the timing of every movement and the camera path.

  5. Add rain to a street

    8 + 8 s · 640 cr
    Add steady rain and wet, reflective pavement to the street. Keep the walkers, their pace and the camera move the same.

  6. Fog through the forest

    10 + 10 s · 800 cr
    Add low morning fog drifting between the trees, softening the background. The hikers, their footsteps and the framing stay untouched.

  7. Swap a held object

    6 + 6 s · 480 cr
    Turn the paper coffee cup in the woman's hand into a glass of orange juice. Keep her hand movement, her face and the kitchen behind her unchanged.

Wan 2.7 modes in the Playcut video studio

In the video studio, pick Wan 2.7 in the model picker, then the mode the card was written for:

  1. Describe for text and multi-shot cards, with a ratio and 2 to 15 seconds.
  2. Animate for talking photos: attach the still, then your WAV or MP3 in the Audio slot. Before recording, time your voiceover to 30 seconds or less.
  3. Reference for cast cards: up to five images, attached in the prompt's order.
  4. Edit for edit cards. If the file is too big, shrink a source clip under 100 MB first.

The studio has one prompt box, so describe exclusions positively. Over the API, a negative prompt reaches Wan on text, reference and edit jobs, not on image to video.

Wan 2.7 costs 40 credits a second at 720p and 60 at 1080p on every mode, so a 10-second 720p clip is 400 credits, held at submit and refunded if the job fails. More on Wan 2.7 modes and credits.

Wan is the only model we run that lip-syncs a still to an audio file you supply. For the same presenter in video after video, the Playcut AI avatar generator saves a reusable actor instead.

FAQ

How do I make a photo speak my audio track in Wan 2.7?

Use image to video: attach the photo plus a WAV or MP3 of 2 to 30 seconds (15 MB at most), and write only the motion and the camera move. Match the clip length to the audio, since a shorter track leaves the rest silent. A 10-second 720p clip costs 400 credits.

What should a Wan 2.7 edit instruction say?

One change and what must stay, such as "make the jacket red; keep the motion and the background". Source and output both run 2 to 10 seconds, and both are billed.

Why doesn't my Wan 2.7 video match my wording?

Alibaba's prompt extension rewrites and expands every Wan prompt with a language model before rendering, so the result is not a literal reading. Keep prompts short and literal, put the must-haves first, and skip exact on-screen text and rapid scene changes.

How do I reference images in a Wan 2.7 prompt?

Write Image 1, Image 2 and so on, with a capital I, a space and no brackets, in attach order. Up to five are allowed. Describe each one next to its tag, as in "the beekeeper in the white veiled hat from Image 1".

How long can a Wan 2.7 clip be?

2 to 15 seconds from text, a photo or references, and 2 to 10 seconds for an edit. A 15-second clip costs 600 credits at 720p or 900 at 1080p.