{
  "motion_tokenizer_config": {
    "in_channels": 3,
    "dim": 160,
    "z_channels": 48,
    "vae_arch": "wan2.2_vae38",
    "dim_mult": [
      1,
      2,
      4,
      4
    ],
    "num_res_blocks": 2,
    "attn_scales": [],
    "temperal_downsample": [
      false,
      true,
      true
    ],
    "dropout": 0.0,
    "embedding_dim": 8,
    "quantizer": "fsq",
    "levels": [
      8,
      5,
      5,
      5,
      5,
      5
    ],
    "num_embeddings": null,
    "beta": 0.25,
    "persistent_quantizer": false,
    "act_embedding_num": 32,
    "qformer_type": "QFormerAdjacentFSingleQ",
    "qformer_depth": 2,
    "qformer_num_heads": 4,
    "connector_out_channels": 1024,
    "name": "WanMotionHybridTokenizer"
  },
  "cross_attention_dim": 3072,
  "projector_hidden_dim": 3072,
  "max_motion_tokens": null
}