{
  "_class_name": "AutoencoderKLConv3D",
  "_diffusers_version": "0.35.0",
  "block_out_channels": [
    128,
    256,
    512,
    1024,
    1024
  ],
  "downsample_match_channel": true,
  "ffactor_spatial": 16,
  "ffactor_temporal": 4,
  "in_channels": 3,
  "latent_channels": 32,
  "layers_per_block": 2,
  "out_channels": 3,
  "sample_size": 256,
  "sample_tsize": 64,
  "scaling_factor": 1.03682,
  "shift_factor": null,
  "upsample_match_channel": true
}