How does Wan's MoE architecture improve performance?

Wan 2.2's Mixture-of-Experts architecture uses 27 billion total parameters with only 14 billion active during inference. This two-expert system maximizes model capacity while maintaining computational efficiency, delivering superior results without increased overhead.

What video capabilities does Wan support?

Wan covers 8 comprehensive tasks: Text-to-Video, Image-to-Video, Video Editing, Text-to-Image, Video-to-Audio, and First&Last-Frame-to-Video. It supports unlimited-length 1080p processing and bilingual text generation with various fonts and effects.

What are the hardware requirements for Wan?

Wan offers flexible deployment options: the 1.3B model requires only 8.19GB VRAM (compatible with consumer GPUs like RTX 4090), while the 14B model delivers SOTA performance. Both variants support multi-GPU acceleration for enhanced productivity.

How does Wan compare to closed-source models?

Wan achieves 84.7%+ comprehensive score on VBench benchmark, often matching or exceeding closed-source models like Sora in specific metrics. It excels in complex dynamics, spatial relationships, and multi-object interactions while remaining fully open-source.

Can Wan be customized for specific use cases?

Absolutely. As a fully open-source platform, Wan can be customized, fine-tuned, and integrated into specific workflows. The comprehensive documentation and model variants enable adaptation for various commercial and research applications.

by Wan AI

Wan AI Video Generator

Versatile video model with strong creative capabilities. Excels at artistic styles, anime, and imaginative content with consistent quality.

Smart Prompt1 Free

Click or drag start image herePNG, JPG up to 10MB

Ratio

Resolution

Time

Leading Open-Source Video Generation Technology

World's First Bilingual Video AI

Breakthrough capability to generate both Chinese and English text directly within video frames with various fonts and effects - the only model supporting dynamic bilingual content.

27B Parameter MoE Architecture

Revolutionary Mixture-of-Experts design with 27 billion total parameters, 14 billion active during inference. Maximizes capacity while maintaining computational efficiency.

Unlimited-Length 1080p Processing

Wan-VAE encodes and decodes unlimited-length 1080p videos without losing temporal information, making it ideal for long-form content and complex video generation tasks.

8-Task Comprehensive Suite

Covers Text-to-Video, Image-to-Video, Video Editing, Text-to-Image, Video-to-Audio, and FLF2V (First&Last-Frame-to-Video) with multi-GPU acceleration support.

VBench Excellence Score 84.7%+

Outstanding performance on authoritative VBench benchmark, excelling in complex dynamics, spatial relationships, and multi-object interactions with SOTA results.

Aesthetic Data Curation

Meticulously curated aesthetic dataset with detailed labels for lighting, composition, contrast, and color tone, enabling precise cinematic style control from noir to commercial-grade.

Wan AI: Frequently Asked Questions

Wan is the first comprehensive open-source video generation suite under Apache 2.0 license, including full source code, models, and documentation. It's also the world's first AI model capable of generating bilingual (Chinese/English) text within videos.

Still have questions? Contact our support team

Explore More AI Video Tools

Text to Video

Transform text prompts into stunning AI videos

Image to Video

Animate your images with AI-powered motion

Reference to Video

Use reference images to guide video style

Watermark Remover

Remove watermarks from videos instantly

Video Upscaler

Enhance video quality up to 4K resolution

Face Swap

Swap faces in videos with AI precision

Experience Open-Source Video AI Excellence -

Unlock the power of the world's most advanced open-source video generation model. Create bilingual content with professional quality and unlimited creative potential.

Generate with Wan AI