Alibaba Drops Wan2.2-S2V: Speech-to-Video Model

Alibaba has unveiled Wan2.2-S2V, an open-source speech-to-video (S2V) AI model that transforms static images and audio clips into lifelike, film-quality animated avatars capable of speaking, singing, and performing.

Key Features

Animation & Performance Capabilities: The model supports a variety of framing options—portrait, bust, and full-body perspectives and enables expressive, natural animations from dialogue to musical performances.

Versatile Character Support: Beyond human avatars, the model supports a diverse range of figures, including cartoons, animals, and stylized characters.

This flexibility makes it suitable for various creative applications, from entertainment content to educational materials.

Model Architecture: The S2V-14B model features 14 billion parameters and utilizes advanced frame processing techniques that compress historical animation frames of arbitrary length into a single, compact latent representation.

This innovative approach significantly reduces computational overhead while enabling stable long-video generation.

Technical Innovation for Long Videos: By leveraging voice-driven local motion and text-guided global control, plus a smart frame-compression method, the model minimizes computational load while maintaining stable long-video generation.

High-Quality Output: The model delivers 480p and 720p resolution outputs, making it adaptable for both quick social-media clips and more refined professional content.

Trained on Industry-Grade Visual Data: Alibaba’s team assembled a large-scale audiovisual dataset tailored to film and television standards. This powering enables precise, story-driven visual expression.

Availability and Access

The Wan2.2-S2V model is completely free to use for both research and commercial applications under the Apache 2.0 license, which permits:

• Commercial use without restrictions.
• Modification and distribution rights.
• Patent rights to users.
• Source code alterations and derivative works.

Download Platforms

The model can be accessed through multiple platforms:

• Hugging Face: Primary distribution platform for model weights.
• GitHub: Source code repository with installation instructions.
• ModelScope: Alibaba Cloud’s open-source community platform.

News Gist

Alibaba has released Wan2.2-S2V, an open-source speech-to-video model that converts static portraits and audio into cinematic avatars.

Available free via GitHub, Hugging Face, and ModelScope, it delivers lifelike animation, realistic lip-sync, and high-quality 480p/720p outputs for global creators.

FAQs

Q1. What is Alibaba Wan2.2-S2V?

It is an open-source speech-to-video AI model that animates static images using audio input, generating film-quality avatars.

Q2. How does Wan2.2-S2V work?

It combines audio-driven local motion with text-guided global control, ensuring realistic lip-sync, facial expressions, and smooth full-body animations.

Q3. What resolutions does it support?

The model produces 480p and 720p videos, suitable for social media, education, and professional content creation.

Q4. Where is Wan2.2-S2V available?

It is free to access on GitHub, Hugging Face, and Alibaba’s ModelScope platform.

Q5. Is there any cost to use the model?

No. Wan2.2-S2V is fully open-source and free of charge.

Q6. Who can benefit from this model?

Developers, educators, researchers, content creators, and filmmakers can use it for digital avatars, storytelling, marketing, and telepresence applications.