Wan 3.0 Prompt Guide: 8 Prompts That Work

What a Wan 3.0 prompt needs

Wan 3.0 is a text-to-video model with native audio. A prompt that only describes a beautiful image leaves half the output to chance. The prompts that work best describe one scene, one action, and the sound that belongs in it.

Keep the scene small enough for 5 or 10 seconds. A clip is not a short film: one location, one subject, one clear motion is usually enough.

The six-part prompt formula

Use this order when a prompt is not working:

1. Subject — who or what is on screen.
2. Action — the single movement that carries the clip.
3. Setting — where it happens and what surrounds the subject.
4. Camera — locked off, handheld, orbit, push-in, tracking.
5. Light — time of day, direction, hardness, color.
6. Sound — ambience, effects, or the absence of sound.

You do not need all six every time, but prompts that include camera and sound tend to produce more controlled results than prompts that only list adjectives.

Eight starter prompts

1. Rainy street at night

A narrow city street after rain, neon signs reflecting in the puddles, one person walking away under a black umbrella, slow handheld camera, soft rain and distant traffic ambience.

2. Product teaser with room tone

A matte black wireless speaker on a concrete plinth, slow 30-degree orbit, hard side light, dust particles in the air, low room tone with a single soft click as the light hits the logo.

3. Quiet morning kitchen

A sunlit kitchen at 7am, steam rising from a ceramic cup, a hand placing toast on a plate, locked-off camera with a slow push-in, birds outside and the quiet hum of a refrigerator.

4. Forest walk after rain

A person in a green rain jacket walking along a wet forest path, camera tracking from behind at shoulder height, mist between the trees, water dripping from leaves, footsteps on soft ground.

5. Coffee pour close-up

A close-up of espresso pouring into a glass cup, crema swirling on the surface, static camera, warm side light, the steady hiss of the machine and a quiet cafe room behind it.

6. City crosswalk at dusk

A crowded crosswalk at dusk, people moving in both directions, camera locked at chest height, cool blue light with warm shop signs, traffic rumble and a crossing signal beeping.

7. Desert train at golden hour

A long freight train crossing an empty desert, camera wide and static, golden hour side light, dust drifting across the tracks, low wind and the distant rhythm of the train.

8. Paper stop-motion desk

A paper cutout bird flapping across a wooden desk, stop-motion feel, top-down camera, soft desk lamp, paper rustle and a quiet room.

How to write sound cues that survive generation

Native audio responds best to sounds that belong to the visible scene. Name two or three sources instead of writing a full sound design brief.

Good cues are concrete: rain on pavement, a refrigerator hum, footsteps on gravel, a distant train. Vague cues such as "epic music" or "cinematic sound" give the model less to work with because they do not point to a physical source.

If you plan to add music later, ask for ambience only. A clean ambience bed is easier to layer under a track than a busy mix.

Common mistakes

Stacking scenes. A prompt that starts in a kitchen and ends on a beach is really two clips. Generate them separately.

Asking for text. Logos, subtitles and signage are unreliable. Add text in the edit.

Expecting 30 seconds. Wan 3.0 in Open Wan renders 5s or 10s. Plan a sequence of clips instead of one long take.

Uploading an image. Wan 3.0 is text to video here. Use Wan 2.7 for image-led shots.

Writing a novel. Long prompts dilute the action. If a sentence does not change what appears on screen or what you hear, cut it.

5 seconds or 10 seconds?

Five seconds is enough for a single gesture, a product reveal or a reaction. Ten seconds gives a camera move time to develop and lets a sound cue breathe.

For a first test, use 5 seconds. Once the composition and motion work, rerun the idea at 10 seconds if the shot needs more room. Both lengths cost 15 credits on Open Wan.

FAQ

Does Wan 3.0 understand camera language?

It responds to common terms such as locked-off, handheld, orbit, push-in and tracking. Use one clear camera instruction instead of several conflicting ones.

Can I include dialogue in the prompt?

You can describe a speaking scene, but treat generated speech as unpredictable. For a scripted voiceover, record it separately and place it over the clip.

Why does the audio not match my prompt?

Audio is generated with the picture, so it follows the visible scene. If the sound is wrong, make the source explicit: name the surface, the room, or the object making the noise.

Where can I try these prompts?

Open the Wan 3.0 model page to see the current settings, or start from the homepage composer and sign in with a new account to use the 20 trial credits.