{
  "1": {
    "class_type": "UNETLoader",
    "_comment": "Z-Image Turbo BF16 transformer",
    "inputs": {
      "unet_name": "z_image_turbo_bf16.safetensors",
      "weight_dtype": "default"
    }
  },
  "2": {
    "class_type": "CLIPLoader",
    "_comment": "Qwen3-4B text encoder — type 'lumina2' routes to z_image.te() for QWEN3_4B model",
    "inputs": {
      "clip_name": "qwen_3_4b.safetensors",
      "type": "lumina2",
      "device": "default"
    }
  },
  "3": {
    "class_type": "VAELoader",
    "_comment": "Z-Image VAE",
    "inputs": {
      "vae_name": "ae.safetensors"
    }
  },
  "4": {
    "class_type": "ModelSamplingAuraFlow",
    "_comment": "Apply ZImage flowmatch sigma schedule (shift=3.0 matches ZImage sampling_settings)",
    "inputs": {
      "model": [
        "1",
        0
      ],
      "shift": 3.0
    }
  },
  "5": {
    "class_type": "LoraLoader",
    "_comment": "Chaerin character LoRA (rank-16, trained with Prodigy on ZIB)",
    "inputs": {
      "model": [
        "4",
        0
      ],
      "clip": [
        "2",
        0
      ],
      "lora_name": "whisper_zit_lora_v1.safetensors",
      "strength_model": 1.0,
      "strength_clip": 1.0
    }
  },
  "6": {
    "class_type": "TextEncodeZImageOmni",
    "_comment": "Positive conditioning — full-body. Edit prompt per shot.",
    "inputs": {
      "clip": [
        "5",
        1
      ],
      "prompt": "cinematic art shot, hands holding microphone close to lips, shallow focus, widescreen horizontal composition, 22 year old Korean girl with soft feminine features, slim elegant frame, cropped black cashmere sweater with collarbone visible,, subtle undereye shadow, baby hairs framing forehead, aegyo sal, low key chiaroscuro with single tungsten, shot on Leica M11 with 50mm Summilux f/1.4, rainy seoul street at night with wet pavement reflections, natural film grain, shallow depth of field, Cinematic widescreen 16:9 landscape orientation.",
      "auto_resize_images": true
    }
  },
  "7": {
    "class_type": "TextEncodeZImageOmni",
    "_comment": "Negative conditioning (empty — ZIT is low-CFG distilled)",
    "inputs": {
      "clip": [
        "5",
        1
      ],
      "prompt": "",
      "auto_resize_images": true
    }
  },
  "8": {
    "class_type": "TextEncodeZImageOmni",
    "_comment": "Face-focused positive for FaceDetailer inpaint pass",
    "inputs": {
      "clip": [
        "5",
        1
      ],
      "prompt": "cinematic art shot, hands holding microphone close to lips, shallow focus, widescreen horizontal composition, 22 year old Korean girl with soft feminine features, slim elegant frame, cropped black cashmere sweater with collarbone visible,, subtle undereye shadow, baby hairs framing forehead, aegyo sal, low key chiaroscuro with single tungsten, shot on Leica M11 with 50mm Summilux f/1.4, rainy seoul street at night with wet pavement reflections, natural film grain, shallow depth of field, Cinematic widescreen 16:9 landscape orientation.",
      "auto_resize_images": true
    }
  },
  "9": {
    "class_type": "EmptySD3LatentImage",
    "_comment": "832x1216 portrait — ZIT uses Flux/16ch latent format (Lumina2 base)",
    "inputs": {
      "width": 1408,
      "height": 768,
      "batch_size": 1
    }
  },
  "10": {
    "class_type": "KSampler",
    "_comment": "Full-body generation — ZIT: 20 steps, cfg=4.0, euler/beta",
    "inputs": {
      "model": [
        "5",
        0
      ],
      "positive": [
        "6",
        0
      ],
      "negative": [
        "7",
        0
      ],
      "latent_image": [
        "9",
        0
      ],
      "seed": 2675622594,
      "steps": 8,
      "cfg": 1.0,
      "sampler_name": "euler_ancestral",
      "scheduler": "beta",
      "denoise": 1.0
    }
  },
  "11": {
    "class_type": "VAEDecode",
    "_comment": "Decode full-body latent → image",
    "inputs": {
      "samples": [
        "10",
        0
      ],
      "vae": [
        "3",
        0
      ]
    }
  },
  "12": {
    "class_type": "UltralyticsDetectorProvider",
    "_comment": "YOLOv8m face bbox detector — models/ultralytics/bbox/face_yolov8m.pt",
    "inputs": {
      "model_name": "bbox/face_yolov8m.pt"
    }
  },
  "13": {
    "class_type": "SAMLoader",
    "_comment": "SAM ViT-B for precise face mask segmentation",
    "inputs": {
      "model_name": "sam_vit_b_01ec64.pth",
      "device_mode": "AUTO"
    }
  },
  "14": {
    "class_type": "FaceDetailer",
    "_comment": "Crop face → 512px → inpaint with LoRA (face large = LoRA tokens dominate) → composite back. denoise=0.45 preserves identity, guide_size=512 ensures face tokens dominate transformer.",
    "inputs": {
      "image": [
        "11",
        0
      ],
      "model": [
        "5",
        0
      ],
      "clip": [
        "5",
        1
      ],
      "vae": [
        "3",
        0
      ],
      "guide_size": 512.0,
      "guide_size_for": true,
      "max_size": 1024.0,
      "seed": 42,
      "steps": 8,
      "cfg": 1.0,
      "sampler_name": "euler_ancestral",
      "scheduler": "beta",
      "positive": [
        "8",
        0
      ],
      "negative": [
        "7",
        0
      ],
      "denoise": 0.45,
      "feather": 5,
      "noise_mask": true,
      "force_inpaint": true,
      "bbox_threshold": 0.5,
      "bbox_dilation": 10,
      "bbox_crop_factor": 3.0,
      "sam_detection_hint": "center-1",
      "sam_dilation": 0,
      "sam_threshold": 0.93,
      "sam_bbox_expansion": 0,
      "sam_mask_hint_threshold": 0.7,
      "sam_mask_hint_use_negative": "False",
      "drop_size": 10,
      "bbox_detector": [
        "12",
        0
      ],
      "wildcard": "",
      "cycle": 1,
      "sam_model_opt": [
        "13",
        0
      ]
    }
  },
  "15": {
    "class_type": "SaveImage",
    "_comment": "Save final face-detailed output",
    "inputs": {
      "images": [
        "14",
        0
      ],
      "filename_prefix": "chaerin_fd"
    }
  }
}