configs/onnx_mp_infer.yaml

models:
  warping_spade:
    name: "WarpingSpadeModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_onnx/warping_spade.onnx"
  motion_extractor:
    name: "MotionExtractorModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_onnx/motion_extractor.onnx"
  landmark:
    name: "LandmarkModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_onnx/landmark.onnx"
  face_analysis:
    name: "MediaPipeFaceModel"
    predict_type: "mp"
  app_feat_extractor:
    name: "AppearanceFeatureExtractorModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_onnx/appearance_feature_extractor.onnx"
  stitching:
    name: "StitchingModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_onnx/stitching.onnx"
  stitching_eye_retarget:
    name: "StitchingModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_onnx/stitching_eye.onnx"
  stitching_lip_retarget:
    name: "StitchingModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_onnx/stitching_lip.onnx"

animal_models:
  warping_spade:
    name: "WarpingSpadeModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_animal_onnx/warping_spade.onnx"
  motion_extractor:
    name: "MotionExtractorModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_animal_onnx/motion_extractor.onnx"
  app_feat_extractor:
    name: "AppearanceFeatureExtractorModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_animal_onnx/appearance_feature_extractor.onnx"
  stitching:
    name: "StitchingModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_animal_onnx/stitching.onnx"
  stitching_eye_retarget:
    name: "StitchingModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_animal_onnx/stitching_eye.onnx"
  stitching_lip_retarget:
    name: "StitchingModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_animal_onnx/stitching_lip.onnx"
  landmark:
    name: "LandmarkModel"
    predict_type: "ort"
    model_path: "./checkpoints/liveportrait_onnx/landmark.onnx"
  face_analysis:
    name: "MediaPipeFaceModel"
    predict_type: "mp"

joyvasa_models:
  motion_model_path: "checkpoints/JoyVASA/motion_generator/motion_generator_hubert_chinese.pt"
  audio_model_path: "checkpoints/chinese-hubert-base"
  motion_template_path: "checkpoints/JoyVASA/motion_template/motion_template.pkl"

crop_params:
  src_dsize: 512
  src_scale: 2.3
  src_vx_ratio: 0.0
  src_vy_ratio: -0.125
  dri_scale: 2.2
  dri_vx_ratio: 0.0
  dri_vy_ratio: -0.1


infer_params:
  flag_crop_driving_video: False
  flag_normalize_lip: True
  flag_source_video_eye_retargeting: False
  flag_video_editing_head_rotation: False
  flag_eye_retargeting: False
  flag_lip_retargeting: False
  flag_stitching: True
  flag_relative_motion: True
  flag_pasteback: True
  flag_do_crop: True
  flag_do_rot: True

  # NOT EXPOERTED PARAMS
  lip_normalize_threshold: 0.03 # threshold for flag_normalize_lip
  source_video_eye_retargeting_threshold: 0.18 # threshold for eyes retargeting if the input is a source video
  driving_smooth_observation_variance: 1e-7 # smooth strength scalar for the animated video when the input is a source video, the larger the number, the smoother the animated video; too much smoothness would result in loss of motion accuracy
  anchor_frame: 0 # TO IMPLEMENT
  mask_crop_path: "./assets/mask_template.png"
  driving_multiplier: 1.0
  animation_region: "all"

  cfg_mode: "incremental"
  cfg_scale: 1.2

  source_max_dim: 1280 # the max dim of height and width of source image
  source_division: 2 # make sure the height and width of source image can be divided by this number