Introduction

Wan 2.2 is a major upgrade from version 2.1, featuring a Mixture-of-Experts (MoE) diffusion architecture, a larger training dataset, and a new 5B hybrid model capable of 720p @ 24fps on a single RTX 4090.
Whether you are working on cinematic text-to-video or image-to-video projects, Wan 2.2 offers smoother motion, richer aesthetics, and more precise prompt adherence.

👉 Read the full Wan 2.2 Prompt Guide here: https://www.wan-ai.co/wan2.2-prompt-guide

Key Upgrades from Wan 2.1

Feature	Wan 2.1	Wan 2.2
Core architecture	Dense diffusion	MoE diffusion with expert hand-off
Training data	Baseline set	+65.6% images, +83.2% videos
Aesthetic control	Basic tags	Cinematic-level controls
Motion fidelity	Moderate	Complex, stable multi-object motion
Model line-up	14B T2V, I2V	14B T2V/I2V + 5B hybrid TI2V

Why this matters:

Sharper frames, fewer artifacts
Smoother and more controllable motion
Lightweight hybrid model for local prototyping

Prompt Writing Formulas

Prompts are the blueprint for your AI video. The more complete and precise they are, the closer the output will match your vision.

Basic Formula

Subject + Scene + Motion

Subject: Main object (person, animal, object, etc.)
Scene: Environment or background details
Motion: Movement of subjects and non-subjects

Example:
A red fox, in a snowy pine forest, trotting slowly while snowflakes fall around it.

Advanced Formula

Subject (Description) + Scene (Description) + Motion (Description) + Aesthetic Control + Stylization

Subject Description: Detailed appearance
Scene Description: Environment adjectives
Motion Description: Speed, amplitude, and effects
Aesthetic Control: Lighting, camera, lens, time of day
Stylization: Artistic styles (e.g., cyberpunk, watercolor)

Example:
A black-haired Miao girl in traditional dress, standing by a river under lanterns, camera dolly-in, warm rim light, Kodak Portra grade, cinematic bokeh.

Image-to-Video Formula

Since the subject and style are defined by the input image, focus on:
Motion + Camera Movement

Motion: What changes in the scene
Camera Movement: e.g., push-in, pan left, fixed camera

Example:
The knight raises his sword slowly, camera pulls back to reveal a vast battlefield.

Advanced Prompt Techniques

Shot Order
Opening shot → Camera motion → Pay-off
Camera Language
Pan, Tilt, Dolly, Orbit, Crane
Motion Modifiers
slow-motion, whip-pan, foreground reeds sway, background mountains fixed
Aesthetic Tags
Lighting: volumetric dusk, neon rim light
Color: teal-and-orange, Kodak Portra
Lens: anamorphic bokeh, 16mm grain
Negative Prompt
bright colors, overexposed, static, blurred details, low quality, extra fingers, malformed limbs, cluttered background

Sample Wan 2.2 Prompts

Neon Drift (Cyberpunk Tracking Shot)
A rainy night in a dense cyberpunk market, neon signs flicker overhead. Camera tracks behind a hooded courier weaving through crowds. Volumetric pink-blue backlight cuts through steam vents, puddles reflect the glow.

Alpine Reveal (Pull Back)
Extreme close-up of a mountaineer’s ice axe biting into frozen rock. Camera pulls back to reveal the climber and a vast sunrise-lit alpine ridge.

Aquatic Ballet (Slow Motion Orbit)
An orca breaches in crystal-clear Arctic waters. Slow 360° orbital shot around the whale, droplets hang suspended. Pastel polar sunset lighting.

Model Variants

Model Type	Model Name	Parameters	Function	Repository
Hybrid Model	Wan2.2-TI2V-5B	5B	Hybrid text-to-video & image-to-video	https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B
Image-to-Video	Wan2.2-I2V-A14B	14B	Converts images to video	https://huggingface.co/Wan-AI/Wan2.2-I2V-A14B
Text-to-Video	Wan2.2-T2V-A14B	14B	Generates video from text	https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B

Conclusion

Wan 2.2 bridges open-source flexibility with cinematic polish. By mastering prompt formulas and camera language, you can achieve stunning, controlled results in both text-to-video and image-to-video generation.

Start writing your best prompts today: https://www.wan-ai.co/wan2.2-prompt-guide

Wan 2.2 – What's New & How to Write Killer Prompts