SCAIL-2: In-Context Character Animation Model Released

SCAIL-2 is an End-to-end control model that animates a reference character via a driving video, without relying on intermediate Skeleton maps or Inpainting masks.

It overcomes the limitations of existing methods, which showed ambiguity in complex motions or were restricted to human-movement input sources. It was trained on 60K motion pairs synthesized using existing models such as SCAIL-Preview, Wan-Animate, and MoCha.

Key features and capabilities are as follows:

• Supports Cross-identity character replacement
• Capable of performing Animal-driving scenarios
• Zero-shot support for advanced control intermediates such as SAM3D-Body mesh rendering