{
    "sample_size": 128,
    "in_channels": 4,
    "out_channels": 4,
    "center_input_sample": false,
    "flip_sin_to_cos": true,
    "freq_shift": 0,
    "down_block_types": [
        "DownBlock2D",
        "CrossAttnDownBlock2D",
        "CrossAttnDownBlock2D"
    ],
    "mid_block_type": "UNetMidBlock2DCrossAttn",
    "up_block_types": [
        "CrossAttnUpBlock2D",
        "CrossAttnUpBlock2D",
        "UpBlock2D"
    ],
    "block_out_channels": [320, 640, 1280],
    "layers_per_block": 2,
    "cross_attention_dim": 2048,
    "transformer_layers_per_block": 10,
    "attention_head_dim": 8,
    "num_attention_heads": 16,
    "use_linear_projection": true
}