{
  "hidden_size": 256,
  "num_attention_heads": 2,
  "attention_head_dim": 128,
  "in_channels": 16,
  "patch_size": 1,
  "joint_attention_dim": 768,
  "pooled_projection_dim": 768,
  "num_double_layers": 3,
  "num_single_layers": 3,
  "mlp_ratio": 4.0,
  "axes_dims_rope": [
    16,
    56,
    56
  ],
  "guidance_embeds": true
}