videoEffect.duration
videoEffect.resolution
videoEffect.ratio
Wan 2.2: How to Access Online & Run Wan 2.2 Locally
Open source meets cinematic precision with the power of Wan 2.2.
What Is Wan 2.2?
What Makes Wan 2.2 Different from Wan 2.1
Wan 2.2 is a next-generation AI video generator developed by Wan AI, a research team affiliated with Alibaba. Designed for text-to-video and image-to-video tasks, it produces high-quality, cinematic videos with faster speed and more realistic motion. Compared to Wan 2.1, this version features a powerful Mixture of Experts (MoE) architecture, enabling smoother generation, better prompt alignment, and stronger visual control.
Wan2.2 Open Source Video Generator
Alibaba Wan 2.2 is fully open-source under the Apache 2.0 license. Developers can freely download, use, and modify the Wan 2.2 AI video generator for both research and commercial projects. This open approach gives creators and engineers full flexibility to build custom Wan video pipelines and integrate the model into their own tools.
Explore Wan 2.2 Examples
Example of Lighting Source
Example of Character Emotion
Example of Stylization
Example of Motion Control
Wan 2.2 AI Video Models
Wan2.2-T2V-A14B: Text-to-Video Model
This model turns text prompts into 5-second videos at 480P and 720P. With strong prompt understanding and motion control, it delivers more accurate and expressive results than previous Wan video versions or competing AI tools.
Wan2.2-I2V-A14B: Image-to-Video Model
Designed for converting static images into dynamic clips, this model supports 480P and 720P output. It reduces camera jitter and supports diverse styles, making it ideal for artistic and storyboard-based applications.
Wan2.2-TI2V-5B: Hybrid Video Generator
The TI2V-5B model supports both text-to-video and image-to-video generation in a single pipeline. It uses a high-compression VAE for fast 720P@24fps video generation, and runs efficiently on a single consumer GPU like an RTX 4090—suitable for research or production.
Key Features of Wan 2.2
Mixture-of-Experts Architecture (MoE)
Wan 2.2 integrates a powerful Mixture-of-Experts (MoE) architecture that enhances model capacity without increasing computational cost. This enables faster, higher-quality video generation across time steps, establishing Wan 2.2 as a technical benchmark in open-source diffusion models.
Large-Scale Training Data
Wan 2.2 is trained with 65.6% more images and 83.2% more videos than its predecessor, Wan 2.1. This substantial increase in training data ensures better motion handling, semantic coherence, and aesthetic consistency in generated videos.
Enhanced Cinematic Aesthetics
With curated datasets labeled for lighting, composition, and color grading, Wan 2.2 excels in producing visually stunning cinematic-style videos, offering creators precise control over their artistic preferences.
Efficient High-Definition Hybrid TI2V
The Wan 2.2 TI2V-5B model supports 720P video generation at 24fps, leveraging advanced VAE compression (16×16×4). Designed for efficiency, it runs on consumer-grade GPUs like the 4090, making high-quality generation accessible to more users.
How to Access Wan 2.2 Online?
Try Wan 2.2 TI2V-5B via Hugging Face Space
You can explore the Wan 2.2 AI video generator directly on Hugging Face Space. The TI2V-5B model supports both text-to-video and image-to-video generation at 720p resolution and 24fps.
Use Wan 2.2 Plus on the Official Website
Visit the official Wan AI site to access Wan 2.2 Plus, a more advanced version offering enhanced tools and preset workflows. It's built for users who want more control and faster output. Note that Wan 2.2 Plus is currently a paid service.
How to Set Up and Run Wan 2.2 Locally
Step 1: Clone the Repository
Begin by cloning the official Wan 2.2 GitHub repository. It contains everything you need to run the model locally.
Step 2: Install Dependencies
Make sure your environment meets the requirements. Install Python packages listed in the repository, and ensure PyTorch is version 2.4.0 or higher for compatibility with Wan 2.2.
Step 3: Download the Model
Select the right model for your use case: T2V-A14B for text-to-video I2V-A14B for image-to-video TI2V-5B for hybrid text+image input, supporting 720p at 24fps You can download and learn more about each model at Hugging Face and ModelScope.
Step 4: Run Video Generation
After setup, you can start generating videos using a text prompt or image. The model will process your input and create high-quality 720p output, all on your local machine—with full control over visual style, movement, and timing.
Get Wan 2.2 ComfyUI Workflow
Wan 2.2 TI2V-5B Workflow for ComfyUI
You can explore a visual workflow to generate videos using Wan 2.2 TI2V-5B directly inside ComfyUI. This setup supports both text-to-video and image-to-video generation. To get started, check out the Wan 2.2 ComfyUI guide, which explains the steps for loading models and running the workflow smoothly.
System Requirements and Setup
Running Wan 2.2 in ComfyUI requires manual installation of specific model files: the diffusion model, VAE, and text encoder. A GPU with at least 8GB VRAM is recommended for smooth generation.
Choose from Three Workflow Types
ComfyUI currently supports three Wan 2.2 workflows: T2V for text-to-video I2V for image-to-video TI2V for hybrid input Each version can be adjusted to fit your prompt, resolution, and frame length preferences—making it flexible for creators, researchers, and developers alike.
How Does Wan 2.2 Work?
1
Efficient MoE Architecture
Wan 2.2 adopts a Mixture of Experts (MoE) design to balance high-quality generation and computational efficiency. During the denoising process, it dynamically switches between two expert models—one specialized for high-noise stages and another for low-noise refinement. This layered structure allows Wan 2.2 to produce more detailed and stable results without increasing memory usage compared to Wan 2.1.
2
Lightweight TI2V-5B for Fast Generation
For users seeking faster performance on consumer hardware, Wan 2.2 includes the TI2V-5B model. It compresses video representations using a high-efficiency VAE, enabling 720p generation at 24fps in under 9 minutes on a single GPU. The model supports both text-to-video and image-to-video tasks within one unified setup.
3
Performance vs. Other AI Video Models
In benchmark comparisons, Wan 2.2 delivers significantly improved accuracy in motion, structure, and prompt alignment. It consistently outperforms leading commercial video models across key evaluation metrics, offering a more realistic and coherent output experience.
Coming Soon: Wan AI Video Generator Integration
You can now try our free AI video generator online at Vidful.ai—no registration required. Full integration with the Wan 2.2 AI video generator is on the way, giving you access to advanced Wan AI models for high-quality, cinematic video creation with fast and efficient performance.
Frequently Asked Questions about Wan 2.2
What is Wan 2.2?
Wan 2.2 is an open-source video generation model, featuring enhanced capabilities like Mixture-of-Experts architecture and cinematic aesthetics.
Is Wan 2.2 really open-source?
Yes, Wan 2.2 is open-sourced, including the TI2V-5B model and supporting components for text-to-video and image-to-video generation.
What hardware do I need to run Wan 2.2?
Wan 2.2 is optimized to run on consumer GPUs like the NVIDIA RTX 4090, delivering high-quality 720P@24fps video generation.
What’s the difference between T2V, I2V, and TI2V in Wan 2.2?
T2V generates videos from text prompts, I2V from static images, and TI2V supports hybrid input for enhanced flexibility.
Can I use Wan 2.2 for commercial projects?
Yes. The open-source license allows for academic and industrial use. Always check the license file for specific permissions.
How does Wan 2.2 compare to commercial models?
Wan 2.2 surpasses leading commercial models in key benchmarks like Wan-Bench 2.0, offering industry-grade output with full transparency.
What resolutions does Wan 2.2 support?
Currently, Wan 2.2 supports video generation at 480P and 720P, with plans for future resolution scalability.