config.json

{
  "_class_name": "ControlNetXSModel",
  "_diffusers_version": "0.22.0.dev0",
  "addition_embed_type": "text_time",
  "base_model_channel_sizes": {
    "dec": [
      [
        2560,
        1280
      ],
      [
        2560,
        1280
      ],
      [
        1920,
        1280
      ],
      [
        1920,
        640
      ],
      [
        1280,
        640
      ],
      [
        960,
        640
      ],
      [
        960,
        320
      ],
      [
        640,
        320
      ],
      [
        640,
        320
      ]
    ],
    "enc": [
      [
        4,
        320
      ],
      [
        320,
        320
      ],
      [
        320,
        320
      ],
      [
        320,
        320
      ],
      [
        320,
        640
      ],
      [
        640,
        640
      ],
      [
        640,
        640
      ],
      [
        640,
        1280
      ],
      [
        1280,
        1280
      ]
    ],
    "mid": [
      [
        1280,
        1280
      ]
    ]
  },
  "block_out_channels": [
    32,
    64,
    128
  ],
  "control_attention_head_dim": 64,
  "control_model_ratio": 0.1,
  "control_scale": 0.95,
  "cross_attention_dim": 2048,
  "down_block_types": [
    "DownBlock2D",
    "CrossAttnDownBlock2D",
    "CrossAttnDownBlock2D"
  ],
  "global_pool_conditions": false,
  "hint_channels": 3,
  "learn_embedding": true,
  "model_channels": 320,
  "out_channels": 4,
  "time_control_scale": 1,
  "time_embedding_dim": 1280,
  "time_embedding_input_dim": 320,
  "transformer_layers_per_block": [
    0,
    2,
    10
  ],
  "up_block_types": [
    "CrossAttnUpBlock2D",
    "CrossAttnUpBlock2D",
    "UpBlock2D"
  ]
}