model_index.json

{
  "_class_name": "MagiHumanPipeline",
  "_diffusers_version": "0.37.0.dev0",
  "transformer": [
    "magi_human",
    "DiTModel"
  ],
  "sr": [
    "magi_human",
    "DiTModel"
  ],
  "audio_vae": [
    "magi_human",
    "SAAudioFeatureExtractor"
  ],
  "text_encoder": [
    "transformers",
    "T5GemmaForCausalLM"
  ],
  "vae": [
    "diffusers",
    "AutoencoderKLWan"
  ],
  "fps": 25,
  "num_inference_steps": 8,
  "video_txt_guidance_scale": 5.0,
  "audio_txt_guidance_scale": 5.0,
  "sr_video_txt_guidance_scale": 3.5,
  "shift": 5.0,
  "cfg_number": 1,
  "sr_cfg_number": 2,
  "noise_value": 220,
  "use_cfg_trick": true,
  "cfg_trick_start_frame": 13,
  "cfg_trick_value": 2.0,
  "using_sde_flag": false,
  "sr_audio_noise_scale": 0.7,
  "t5_gemma_target_length": 640,
  "vae_stride": [4, 16, 16],
  "z_dim": 48,
  "patch_size": [1, 2, 2],
  "data_proxy": {
    "t_patch_size": 1,
    "patch_size": 2,
    "frame_receptive_field": 11,
    "spatial_rope_interpolation": "extra",
    "ref_audio_offset": 1000,
    "text_offset": 0,
    "coords_style": "v2"
  }
}