Skip to content

MotionCLIP Text Encode

Documentation

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

The MotionCLIPTextEncode node is designed to encode textual descriptions into a format suitable for motion generation tasks, leveraging a CLIP model and motion data to produce motion conditioning. This process facilitates the creation of motion sequences that are aligned with the given textual descriptions.

Input types

Required

  • md_clip
    • The md_clip parameter represents a CLIP model tailored for motion data, crucial for interpreting and encoding the textual description in the context of motion.
    • Comfy dtype: MD_CLIP
    • Python dtype: MD_CLIP
  • motion_data
    • The motion_data parameter contains the motion information necessary for the encoding process, providing the context in which the text is to be interpreted.
    • Comfy dtype: MOTION_DATA
    • Python dtype: MOTION_DATA
  • text
    • The text parameter is the textual description to be encoded, serving as the input for generating motion-aligned sequences.
    • Comfy dtype: STRING
    • Python dtype: str

Output types

  • md_conditioning
    • Comfy dtype: MD_CONDITIONING
    • The output is motion conditioning data, which is used to guide the generation of motion sequences that correspond to the input text.
    • Python dtype: MD_CONDITIONING

Usage tips

  • Infra type: GPU
  • Common nodes: unknown

Source code

class MotionCLIPTextEncode:
    @classmethod
    def INPUT_TYPES(s):
        return {
            "required": {
                "md_clip": ("MD_CLIP", ),
                "motion_data": ("MOTION_DATA", ),
                "text": ("STRING", {"default": "a person performs a cartwheel" ,"multiline": True})
            },
        }

    RETURN_TYPES = ("MD_CONDITIONING",)
    CATEGORY = "MotionDiff"
    FUNCTION = "encode_text"

    def encode_text(self, md_clip, motion_data, text):
        return (md_clip(text, motion_data), )