This is an eye-closed smile motion lora trained on 5 anime clips with animatediff v2 motion module.
It's not stable, I'll release some later versions to improve the quality, but currently for better movement, using a weight above 1.2 is recommended.
The prompts would affect the motion badly, do not use other prompts not shown in demo videos like "upper body, pants, breast, pants, white sleeves". Just focus on the face, hair prompts, or upper body clothes. realcartoon2dv9 is used in demos.
The comfyUI workflow is embedded in images, save and drag the images into comfyUI to load them.