Wan 2.2

Wan 2.2: Leading AI Video Generation Model

Wan 2.2 is an AI creative platform developed by Alibaba, designed to lower the barriers to creative work through artificial intelligence. It provides functionalities like text-to-image, image-to-image, text-to-video, image-to-video, and image editing.

What is Wan 2.2?

Wan 2.2 is a significant upgrade to Alibaba's visual generative models, now open-sourced. This release offers enhanced capabilities, better performance, and superior visual quality, focusing on incorporating technical innovations like MoE architecture, data scaling, cinematic aesthetics, and efficient high-definition hybrid TI2V.

Key Features and Capabilities:

• Cinematic Vision Control: Achieves professional cinematic narratives through fine-grained control over lighting, color, and composition.
• Sweeping Motion: Effortlessly recreates various complex motions with enhanced fluidity and control.
• Precise Prompt Following: Better understands and executes prompts for complex scenes and multi-object generation.
• Wan Box Project: Integrates various creation tasks, including image and video generation and editing, within a single interface.

How does Wan 2.2 work?

Wan 2.2 incorporates several technical innovations:

• MoE Architecture: Introduces a Mixture-of-Experts (MoE) architecture into video diffusion models. This separates the denoising process across timesteps using specialized expert models, increasing overall model capacity while maintaining computational efficiency. The A14B model series employs a two-expert design, using a high-noise expert for early stages and a low-noise expert for refining video details.
• Data Scaling: Trained on significantly larger datasets compared to Wan 2.1 (+65.6% more images and +83.2% more videos), enhancing the model's generalization across motions, semantics, and aesthetics.
• Cinematic Aesthetics: Incorporates curated aesthetic data with fine-grained labels for lighting, composition, and color, enabling more precise and controllable cinematic style generation.
• Efficient High-Definition Hybrid TI2V: Open-sources a 5B model built with the advanced Wan2.2-VAE, achieving a compression ratio of 16×16×4. This model supports both text-to-video and image-to-video generation at 720P resolution with 24fps and can run on consumer-grade graphics cards like the 4090.

Open Source Availability

Wan 2.2 is open-sourced, offering powerful capabilities, better performance, and superior visual quality. The open-source release includes:

• Wan2.2-T2V-A14B: Supports generating 5-second videos at 480P and 720P resolutions, surpassing leading commercial models in key evaluation dimensions.
• Wan2.2-I2V-A14B: Designed for image-to-video generation, achieving more stable video synthesis and enhanced support for diverse stylized scenes.
• Wan2.2-TI2V-5B: Supports both text-to-video and image-to-video generation at 720P resolution with 24fps, capable of running on a single consumer-grade GPU.

Wan Box: All in Wan, Create Anything

Wan Box allows users to initiate various creative tasks, including image generation, video generation, and video editing. It offers flexible video clip editing using a Time Line to splice clips and perform further generation.

Why is Wan 2.2 important?

Wan 2.2 lowers the barrier to entry for AI-driven creative video generation, enabling both industrial and academic sectors to leverage its advanced capabilities. Its open-source nature fosters collaboration and innovation in the field.

Examples of Wan 2.2 in Action:

• Cinematic Scenes: Create stunning videos with fine-grained control over cinematic elements. Examples include a young man in a sunlit forest, a train moving across a stage bathed in spotlights, and a person on an escalator with mirrored reflections.
• Dynamic Motion: Generate videos featuring complex and fluid motion, such as hip-hop dancing, street parkour, and figure skating.
• Imaginative Scenarios: Produce unique and visually striking scenes, such as a woman blowing a bubble with a miniature aquarium inside and a woman using a garden hose that sprouts colorful flowers.

Comparisons to State-of-the-Art Models

Wan 2.2 has been compared to leading closed-source commercial models on Wan-Bench 2.0, demonstrating superior performance across multiple critical dimensions. This highlights its advanced capabilities and positions it as a leader in the field of AI video generation.

Where can I use Wan 2.2?

Wan 2.2 is suitable for various applications, including:

• Content creation for social media
• Marketing and advertising
• Educational videos
• Artistic expression
• Research and development in AI video generation

How to get started with Wan 2.2?

Visit the official Wan website and access the open-source models. You can experiment with the various generation modes, including text-to-video and image-to-video, to create your own AI-powered videos.

In summary, Wan 2.2 stands as a groundbreaking AI video generation model, offering a blend of advanced technology, creative flexibility, and accessibility through its open-source release. It's set to empower both professionals and enthusiasts in the creation of visually stunning and dynamic video content.