Wan 2.2, developed by Alibaba's Tongyi Lab, is an upgraded open-source AI video generation model that builds on Wan 2.1. Here are the key new features and improvements:
Native 1080p Resolution: Wan 2.2 supports full HD (1080p) video output, delivering crisper details and sharper text without requiring upscaling, unlike Wan 2.1’s focus on 480p and 720p.
Mixture of Experts (MoE) Architecture: It introduces an innovative MoE model with high-noise and low-noise expert models, optimizing denoising for higher-quality video content. This enhances performance and scalability.
Enhanced VACE 2.0: The Video Animation Control Engine (VACE) is upgraded, offering advanced camera trajectory controls (pans, zooms, focus pulls), subject locking for precise tracking, and background stabilization for smoother shots.
Improved LoRA Personalization: Wan 2.2 features a "few-shot" LoRA (Low-Rank Adaptation) pipeline, allowing style adaptation with just 10–20 images. It includes visual LoRA sliders for intuitive style blending, ideal for custom characters or branded content.
Volumetric Effects: Supports realistic visual effects like fire, smoke, and dynamic lighting, enhancing visual quality for complex scenes.
Cinematic Aesthetic Control: Offers professional-grade control over lighting, color, and composition, aligning with film industry standards.
Efficient Compression: The 5B parameter Text+Image-to-Video (TI2V-5B) model uses a high-compression VAE, optimizing memory usage while maintaining quality.
Multimodal Generation: Supports Text-to-Video (T2V-A14B), Image-to-Video (I2V-A14B), and Text+Image-to-Video (TI2V-5B) modes, all at up to 720p, with plans for 1080p.
Bilingual Prompting: Continues robust support for English and Chinese prompts, including accurate in-frame text rendering.
Leaner Architecture: Expected to have around 10B parameters (down from Wan 2.1’s 14B), making it more efficient for consumer hardware and cloud platforms.
Open-Source Accessibility: Licensed under Apache 2.0 for commercial use, ensuring broad accessibility without vendor lock-in.
Wan 2.2 was released on July 28, 2025, and is available on platforms like Atlabs AI, with free trials and cloud-based workflows for users without high-end GPUs. It's performance is comparable with top models like Veo2, Kling 2.1, Seedance 1.0 while cheaper to use.
Note: To create longer videos (>10s), you can use Atlabs storyboard to AI video pipeline. It uses a mixture of creative AI agents and editing suite for full creative control.
