WanX AI

WanX 2.1 is an advanced multimodal AI model designed to transform text inputs into high-quality videos and images, redefining AI-driven visual content creation. Launched as an evolution of the Tongyi Wanxiang model (debuted in July 2023), WanX 2.1 excels in generating realistic visuals with a focus on complex movements, enhanced pixel quality, and adherence to physical rules.

It leverages a proprietary Variational Autoencoder (VAE) and Denoising Diffusion Transformer (DiT) framework, combined with a full space-time attention mechanism, to ensure precise spatial-temporal relationships and lifelike dynamics. This makes it particularly adept at handling challenging scenarios like figure skating, swimming, or diving, maintaining body coordination and realistic motion trajectories.

The model supports both Chinese and English text inputs, a groundbreaking feature that broadens its appeal for global creative industries such as advertising, short video production, and education. WanX 2.1 tops the VBench leaderboard with an overall score of 84.7%, leading in categories like dynamic degree, spatial relationships, and multi-object interactions. It can generate 1080p videos in as little as 15 seconds per minute of content, offering over 100 artistic styles (e.g., cyberpunk, oil painting) for customization. Currently available for free on Alibaba Cloud’s Model Studio and its official Chinese website, WanX 2.1 is set to be fully open-sourced in Q2 2025, including its training datasets and developer toolkit, promising to democratize access and spur innovation in AI video generation.

In essence, WanX 2.1 is a cutting-edge tool that blends speed, quality, and versatility, positioning Alibaba Cloud as a major player in the generative AI landscape, with applications spanning marketing, entertainment, and beyond.