Skip to content
Log inSign up free

Prompting guide

How to prompt Wan 2.2

How to prompt Wan 2.2 on Mage: a quick start, a fill-in template, and expert tips for cinematic tags, motion, start frames, styles, and negative prompts.

Try Wan Video v2.2Wan Video v2.2 model page

Kite on a hilltop
Made with Wan Video v2.2Kite on a hilltop

Quick start

Your first great Wan 2.2 prompt

Everything a beginner needs in one minute: what to know first, the rules that matter most, a template to fill in, and prompts to copy.

What to know first

Wan 2.2 makes short, silent clips from a prompt or from a start image. It learned from footage labeled for light, framing, and color, so it answers to the words a cinematographer uses. Open with up to four of those tags, then describe the subject, tell the action as it unfolds, and give the background something that moves. Leave sound out: the clip has none.

The rules that matter most

  1. Lead with up to four cinematic tags

    Pick from time of day, light, shot size, angle, and composition: "Side lighting, medium shot, sunrise time, warm colors." More than four dilutes them.

  2. Tell the action from start to finish

    "She reaches for the cup, lifts it, and blows across the surface." A subject that only stands there gives a clip that barely moves.

  3. Give the background motion

    Clouds drifting, leaves stirring, traffic passing. Even a small movement behind the subject keeps the shot alive.

  4. Leave out sound and mood sentences

    No dialogue, music, or effects: the clip is silent. No lines like "the scene is full of life": show what makes it so.

  5. From an image, write only the motion

    With a First Frame, the picture already sets the look. Say what moves and what the camera does, in under 100 words.

Fill in this template

Replace each part in square brackets with your own idea, and leave out a part you do not need.

Template
[Up to four cinematic tags]. [Subject, with looks, expression, and pose] [does the action, from start to finish]. [Background, with something in it moving].
The template, filled in
Side lighting, medium shot, sunrise time, warm colors. A fisherman in his sixties with a gray beard and a yellow oilskin jacket stands in a small wooden boat, hauls a dripping net over the side hand over hand, then holds up a silver fish and grins. Gulls wheel over the calm bay under a clear blue sky, and the boat rocks gently on the swell.
  • [Up to four cinematic tags] Time, light, shot size, angle, or composition, comma-separated: Side lighting, medium shot, sunrise time, warm colors.
  • [Subject, with looks, expression, and pose] Who or what, with the details that show: a fisherman in his sixties with a gray beard and a yellow oilskin jacket, standing in a small wooden boat.
  • [does the action, from start to finish] The action as a process: hauls a dripping net over the side hand over hand, then holds up a silver fish and grins.
  • [Background, with something in it moving] The setting, and its motion: gulls wheel over the calm bay under a clear blue sky.

Prompts to start from

  • A portrait in motion

    Practical lighting, warm colors, night time, medium close-up shot. A woman in her thirties with cropped black hair, a mustard knit cardigan and round tortoiseshell glasses sits at a cluttered desk under a green-shaded lamp. She leans in to read a letter, frowns, then breaks into a slow smile and presses the page to her chest. Behind her, rain streaks the dark window and the reflection of the lamp trembles in the glass.

    Why it works: Four tags set the light and the framing, the action has a beginning, a turn, and an end, and the rain gives the background its own motion.

  • A landscape with no people

    Aerial, extreme wide shot, dawn time, soft lighting. Mist rolls slowly across terraced rice fields that step down a green hillside, the flooded terraces mirroring a pale pink sky. A narrow river winds along the valley floor, and a flock of white egrets lifts off the water and glides across the frame.

    Why it works: No people were asked for, so none are added; the mist, the river, and the birds supply the motion.

  • From a first frame

    The girl in the yellow raincoat jumps into the puddle with both feet, water splashing up around her boots, then looks up and laughs. The camera pulls back slowly.

    Why it works: Upload the still as the First Frame. The prompt names the subject as the picture shows her and says only what moves.

  • A stylized clip

    Claymation. A chubby clay chef in a tall white hat flips a pancake in a tiny frying pan; it spins high into the air, and he catches it on a plate with a proud little bow. Behind him, steam puffs from a row of clay pots on a miniature stove.

    Why it works: The named style comes first and the cinematic tags are left out, as they should be for anything that is not a photograph.

Advanced

Advanced Wan 2.2 prompting

Techniques for more control once the basics work.

Wan Video v2.2 or Lightning

Wan Video v2.2 gives the better result. Wan Video v2.2 Lightning is faster and cheaper, for drafts. Both read the same prompt, so work the motion out on Lightning and render the keeper on the standard model without changing a word.

Structure: tags, subject, action, background

A text-to-video prompt has four parts, in this order: up to four cinematic tags, comma-separated; the subject, with its looks, expression, count, and pose; the action, told as a process from start to finish; and the background, with something in it moving.

Aim for 60 to 200 words, in English. Wan 2.2 reads up to roughly 350 words, but the strongest prompts stay well under that.

  • Add detail to the subjects you named, but never add subjects you did not ask for: a landscape stays empty of people.
  • If you have no action in mind, add one, even a small one, and give the background motion too.
  • Write "a clear blue sky", not just "sky": a bare sky tends to blow out to white.
  • Leave out literary atmosphere sentences ("the scene is full of tension") and describe what is visible instead.

Cinematic tags

Pick up to four tags from the menu below; you need not use every kind. With no tags, Wan 2.2 tends toward daytime, a medium or wide shot, and a centered composition. Say where the light comes from when it matters: "Practical lighting from a desk lamp".

WriteWhat it does
Day time, Night time, Dawn time, Sunrise timeThe time of day.
Daylight, Moonlight, Firelight, Practical lighting, Fluorescent lighting, Overcast lighting, Sunny lighting, Artificial lightingThe light source. Practical lighting means lamps and screens visible in the scene.
Soft lighting, Hard lightingGentle, diffused light or crisp light with sharp shadows.
Top lighting, Side lighting, Underlighting, Edge lightingWhere the light hits from. Edge lighting outlines the subject against the background.
Warm colors, Cool colors, Mixed colorsThe palette. Add low saturation, high saturation, or high contrast as needed.
Extreme close-up, Close-up, Medium close-up, Medium shot, Medium wide shot, Wide shot, Extreme wide shotHow much of the subject fills the frame.
Over-the-shoulder, Low angle, High angle, Dutch angle, Aerial, OverheadWhere the camera looks from. Skip this tag when the prompt already has a camera move.
Center, Balanced, Left-heavy, Right-heavy, Symmetrical, Short-side compositionWhere the subject sits in the frame. Short-side leaves the empty space behind the subject instead of in front.

Camera and motion

Every camera move says something. Push in for intimacy or tension, pull back to reveal, track to travel with the subject, orbit to make something feel important, and hold still for calm. Keep an orbit to a short arc, under about 45 degrees, or the space starts to warp.

A clip is one continuous shot. Tell the action as it unfolds ("she reaches for the cup, lifts it, and blows across the surface") and keep it to what fits in a few seconds: no cuts, no second scene.

WriteWhat it does
long focusA telephoto look: a compressed, blurred background.
wide-angleMore of the setting in frame, with depth.
ultra-wide fisheyeA strongly curved, bulging view.
tilt-shiftA real scene that looks like a miniature model.
the camera pans upThe view tilts up to reveal what is above.
moves from left to rightThe camera slides sideways across the scene.

Starting from an image

Upload a still as the First Frame and the clip opens on it, in its shape. Then the prompt is about motion only: what the subject does and what the camera does. Leave out everything the picture already shows, and keep it under 100 words.

Name the subject the way the picture shows it. If all you have is an action, like "dancing", write "the girl in the yellow dress dances".

  • Always give an action, even a small one: a head turn, steam rising, leaves stirring. A frame with no described motion tends to come out nearly still.
  • Keep the camera moves you want, and say their direction: "the camera pans up", "moves slowly from left to right".
  • Wan 2.2 takes no Characters, References, or @handles, and no last frame. To keep a character the same across clips, start each one from a still of them, made on Guava or Mango for example, and describe them the same way each time.

Styles

Put a named style first, before anything else. For a style that is not a photograph, such as a 2D illustration or anime, leave the cinematic tags out: they pull the clip back toward live action.

WriteWhat it does
feltSoft wool-felt figures and sets.
3D cartoonRounded, glossy animated characters.
pixel artBlocky retro game graphics.
puppet animationHandmade puppets on a small set.
claymationStop-motion clay figures, fingerprints and all.
black-and-white animationClassic monochrome cartoon.
tilt-shift photographyA real place shot to look like a toy model.
time-lapseHours compressed into seconds.

A city that looks like a model

Tilt-shift photography. A high-angle view of a busy harbor at golden hour: container cranes swing slowly, tiny trucks crawl along the quay, and a tugboat pushes a cargo ship away from the dock, its wake spreading in a neat white fan across the water.

Why it works: The style leads, the high angle sells the miniature, and several small things move at once.

The negative prompt

By default, Mage adds Wan 2.2's standard list of flaws to avoid: garish colors, overexposure, a frozen frame, blurry detail, subtitles, a painted look, a grayish tone, low quality, compression artifacts, extra or fused fingers, badly drawn hands and faces, extra legs, a cluttered or crowded background, and people walking backwards. Anything you put in the Negative Prompt setting is added after it.

So add only what is specific to this shot, such as "umbrellas, wet pavement" for a dry street, or "text, logos" for clean signs.

Sound and text

Wan 2.2 clips are silent. Never write dialogue, voices, music, or effects; add audio afterwards, for example with Seed Audio.

Text is unreliable. A short word in quotes sometimes holds, like a name tag reading "Mia"; add anything that must be readable in an editor.

Common mistakes

Avoid this, do this instead

  • Avoid: A clip that barely moves.

    A lighthouse on a cliff at sunset.

    Instead: The action as a process, plus motion in the background.

    Wide shot, sunset time, warm colors, edge lighting. A white lighthouse stands on a grassy cliff as its lamp blinks on and begins to turn. Waves roll in and break against the rocks below, and a line of seabirds glides past the tower.

  • Avoid: Mood adjectives instead of detail.

    Instead: Concrete looks, actions, and light.

  • Avoid: More than four cinematic tags.

    Instead: The four that matter most.

  • Avoid: Sound, dialogue, or music.

    Instead: Nothing: the clip is silent.

  • Avoid: Describing a first frame all over again.

    Instead: Only the motion and the camera.

  • Avoid: Fast cuts or several scenes.

    Instead: One continuous shot.

  • Avoid: People you did not ask for in a landscape.

    Instead: Only the subjects you named, with motion from nature.

FAQ

Wan 2.2 prompting questions

How long should a Wan 2.2 prompt be?
60 to 200 words from text, and under 100 from a start image. Wan 2.2 reads up to roughly 350 words, but tight prompts in the tags, subject, action, background order work best.
Why is my Wan 2.2 clip barely moving?
The prompt probably describes a scene but no action. Tell what happens from start to finish, and give the background something that moves, such as clouds, water, or leaves.
Does Wan 2.2 make sound?
No. Wan 2.2 clips are silent. Add music, voices, or effects afterwards, for example with Seed Audio.
Does Wan 2.2 support negative prompts?
Yes. Mage adds the standard negative list for you; put only what is specific to your shot in the Negative Prompt setting.
How do I keep a character consistent in Wan 2.2?
Wan 2.2 takes no Characters or references. Start each clip from a still of the character as the First Frame, and describe them the same way in every prompt.
Which tags does Wan 2.2 understand?
Time of day, light source, light quality and angle, color, shot size, camera angle, and composition, such as "Side lighting, medium shot, sunrise time, warm colors". Use four at most.
Should I use Wan Video v2.2 or Lightning?
Lightning for fast, cheap drafts, and the standard Wan Video v2.2 for the final. The same prompt works on both.

Make your first video with Wan 2.2

New accounts get 300 free Gems, enough to try Wan Video v2.2. No install, no credit card.

Try Wan Video v2.2