The Core Formula: Six Building Blocks
Wan responds best to prompts assembled in a consistent order. The structure recommended by Alibaba's own documentation — and validated across thousands of community generations — is:
- Subject — who or what the video is about ('a young chef', 'a matte black sports car')
- Scene — where it happens ('in a neon-lit ramen shop', 'on a windswept cliff at dawn')
- Motion — what physically happens ('kneads dough rhythmically', 'drifts around a corner, tires smoking')
- Camera language — how the camera behaves ('slow dolly in', 'orbiting crane shot')
- Atmosphere — light, weather, mood ('golden hour haze', 'moody overcast with drifting fog')
- Style — the visual register ('cinematic photorealistic', '90s cel anime', 'documentary handheld')
Ideal Prompt Length: 80–120 Words
Short prompts ('a cat playing piano') leave every creative decision to the model — results are generic and inconsistent. The sweet spot is roughly 80–120 words: enough to lock subject, motion and mood, but not so dense that instructions conflict.
If you must write short prompts, prioritize motion and camera language over adjectives — those two blocks have the largest impact on output quality.
Image-to-Video Prompts: Describe Motion Only
When animating a photo, your image already defines the subject, scene and style. Adding them again confuses the model. The I2V formula is simply: Motion + Camera movement.
Two rules matter more than anything else in I2V: only describe elements actually visible in the image (the model will not invent new objects), and keep camera moves minimal — a slight zoom or gentle pan outperforms dramatic tracking shots.
One Action per Clip
Wan renders 5–15 second clips. Cramming three actions into one prompt produces morphing transitions between them. Choose the single most important action and let the atmosphere carry the rest.
Need a sequence? Generate multiple clips and cut them together — you'll get cleaner results and full editorial control.
Negative Prompts: Your Quality Insurance
A short negative prompt list dramatically reduces artifacts. Our standard baseline: 'morphing, warping, distorted face, extra fingers, blurry, flickering, low quality'. For the full vocabulary see our Negative Prompts dictionary.