{
  "denoiser": {
    "name": "LatentSeqMMFlowModel",
    "args": {
      "q_token_length": 8192,
      "in_channels": 16,
      "cam_channels": 5,
      "model_channels": 1024,
      "cond_channels": 1280,
      "cond2_channels": 128,
      "num_blocks": 24,
      "num_refiner_blocks": 2,
      "num_heads": 16,
      "mlp_ratio": 4
    }
  },
  "decoder": {
    "name": "OctreeProbabilityFixedlenDecoder",
    "args": {
      "model_channels": 1024,
      "cond_channels": 16,
      "num_blocks": 4,
      "num_heads": 16,
      "mlp_ratio": 4
    }
  },
  "decoder_gs": {
    "name": "GaussianFixedlenDecoder",
    "args": {
      "in_channels": 3,
      "model_channels": 1024,
      "cond_channels": 16,
      "num_blocks": 16,
      "num_heads": 16,
      "mlp_ratio": 4,
      "representation_config": {
        "num_gaussians": 32,
        "lr": {
          "_xyz": 1.0,
          "_features_dc": 1.0,
          "_opacity": 1.0,
          "_scaling": 1.0,
          "_rotation": 0.1
        },
        "perturbe_size": 1.5,
        "3d_filter_kernel_size": 0.0009,
        "scaling_bias": 0.004,
        "opacity_bias": 0.1
      }
    }
  },
  "image_size": 1024,
  "max_voxel_level": 8,
  "image_encoders": {
    "dinov3": "facebook/dinov3-vith16plus-pretrain-lvd1689m",
    "flux2_vae": "black-forest-labs/FLUX.2-dev"
  }
}