config.json

{
  "in_channels": 3,
  "out_channels": 3,
  "down_block_types": [
    "ResBlock", "ResBlock", "DownSample", 
    "ResBlock", "AttnResBlock", "ResBlock", "DownSample",
    "ResBlock", "ResBlock", "DownSample",
    "ResBlock", "ResBlock"
  ],
  "up_block_types": [
    "ResBlock", "ResBlock",
    "ResBlock", "ResBlock", "DownSample",
    "ResBlock", "AttnResBlock", "ResBlock", "DownSample",
    "ResBlock", "ResBlock", "DownSample"
  ],
  "block_out_channels": [
    128, 256, 256, 256
  ],
  "layers_per_block": 2,
  "norm_num_groups": 32,
  "cross_attention_dim": null,
  "attention_head_dim": 8,
  "resnet_time_scale_shift": "default",
  "class_embed_type": null,
  "num_class_embeds": null,
  "use_linear_projection": false,
  "only_cross_attention": false,
  "mid_block_scale_factor": 1,
  "out_bias": true,
  "flip_sin_to_cos": true,
  "freq_shift": 0
}