Wan

Wan constitutes an artificial intelligence platform oriented towards visual content creation through generative models. The tool offers image and video generation capabilities from different input types, including text, reference images, and audio.

In the video generation domain, the platform provides several specialized models. The Text to Video function allows creating videos from textual descriptions with interpretation of cinematic instructions. Image to Video transforms static images into animated sequences while maintaining coherence with the original visual content. Reference to Video enables character transfer from reference videos to new scenes, with support for human or human-like figures and capability to maintain consistency in appearance and voice. Speech to Video generates character videos from an image and an audio clip, using audio to control facial expressions and body movements across different character types.

Image generation is performed through the Text to Image module, which processes textual descriptions to produce visual content with different aesthetic styles. The system incorporates instruction interpretation for generating images consistent with provided specifications.

WanBox functions as a workspace where image generation, video creation, and editing tasks are initiated. The platform includes a project system with timeline that facilitates clip assembly, video editing, and additional generations on existing material.

The platform is structured as a web service accessible through browsers, providing an interface for generation parameter configuration and result visualization.

Wan offers open-source models that include character animation and replacement capabilities, audio-driven video generation, unified models for video creation and editing, and sequence generation from initial and final frames. The models employ architectures based on diffusion transformers and techniques such as Mixture of Experts for processing.