MotionDiff Loader

Documentation

• Class name: MotionDiffLoader
• Category: MotionDiff
• Output node: False

The MotionDiffLoader node is designed to load and initialize the Motion Diffusion Model (MDM) and its associated CLIP wrapper based on a specified model dataset. This node plays a crucial role in preparing the motion generation models for subsequent processing or inference tasks, ensuring they are correctly configured with the necessary dataset information.

Input types

Required

• model_dataset
  • Specifies the dataset model to be loaded. This selection determines the configuration of the Motion Diffusion Model and its CLIP wrapper, impacting the behavior and performance of motion generation.
  • Comfy dtype: COMBO[STRING]
  • Python dtype: str

Output types

• md_model
  • Comfy dtype: MD_MODEL
  • Returns a wrapped instance of the Motion Diffusion Model, ready for motion generation tasks.
  • Python dtype: MotionDiffModelWrapper
• md_clip
  • Comfy dtype: MD_CLIP
  • Returns a CLIP wrapper configured for the loaded Motion Diffusion Model, facilitating text-conditioned motion generation.
  • Python dtype: MotionDiffCLIPWrapper

Usage tips

• Infra type: GPU
• Common nodes: unknown

Source code

class MotionDiffLoader:
    @classmethod
    def INPUT_TYPES(s):
        global model_dataset_dict
        model_dataset_dict = get_model_dataset_dict()
        return {
            "required": {
                "model_dataset": (
                    list(model_dataset_dict.keys()), 
                    { "default": "-human_ml3d" }
                )
            },
        }

    RETURN_TYPES = ("MD_MODEL", "MD_CLIP")
    CATEGORY = "MotionDiff"
    FUNCTION = "load_mdm"

    def load_mdm(self, model_dataset):
        global model_dataset_dict
        if model_dataset_dict is None:
            model_dataset_dict = get_model_dataset_dict() #In case of API users
        model_config = model_dataset_dict[model_dataset]()
        mdm = create_mdm_model(model_config)
        return (MotionDiffModelWrapper(mdm, dataset=model_config.dataset), MotionDiffCLIPWrapper(mdm))