WAN2.1 Lynx Последовательный Персонаж-KJ

Wan2.1 Lynx is a text-to-video (T2V) model variant under the Wan2.1 architecture. Its main feature is improved consistency in preserving the identity and core appearance of the input subject during automatic video generation. The model aims to retain facial features and prominent visual elements of the given character throughout the generated sequence, making it suitable for tasks such as personalized animation, virtual avatars, or digital twin applications.

Lynx incorporates additional feature adapters, such as an ArcFace-driven ID-adapter, to extract and align subject-specific information. This helps reduce visual drift in faces and key features during generation, resulting in more stable and visually consistent animations.

The model can be integrated into ComfyUI workflows and supports typical high-resolution AIGC creation pipelines.