This LoRA is designed for anime character head-turning motion enhancement. It works best with WanT2V-14B, and also performs reasonably well with AniWan. When used together with CausVid or Acc, it may cause blocky noise artifacts during head-turning frames—see the comparison images for visual differences. For optimal results, it is recommended to apply a VACE to improve the quality and consistency of the first and last frames.Simple PromptAnimeTH. When two frames are provided, we analyze the difference to describe the head-turning motion in detail, including direction (e.g., left/right/diagonal), angle (15–30° slight / 45–90° moderate / 90–180° large), speed (graceful / natural / sharp), and hair dynamics—such as flutter direction, volume, and texture (silky / fluffy / elastic curls). With a single image, we infer a natural motion consistent with the character’s pose and style, keeping the flow and hair movement elegant and expressive. The tone and vocabulary align with the image style—be it anime, cinematic realism, or stylized 2D.

Files with SHA256: 0bbdfce0086653b9a4dd7057702b8334ba4d5d2d8a6cbe393093697337ecb633