{
  "1": {
    "class_type": "UNETLoader",
    "inputs": {
      "unet_name": "minimax_h3_ref2va_pruned_int8_convrot.safetensors",
      "weight_dtype": "default"
    }
  },
  "m0": {
    "class_type": "MiniMaxH3TurboLoRA",
    "inputs": {
      "lora_name": "minimax_h3_fl2v_lightx2v_turbo_8step_v1.0_comfy.safetensors",
      "strength": 1.0,
      "low_vram": false,
      "model": [
        "1",
        0
      ]
    },
    "_comment": "lightx2v Minimax-h3 Turbo **8step v1.0** 단독 @1.0 (2026-08-12 `t_4e40bb6cf50d` · CEO 승격 지시). 구 `ckpt850`(4step) 에서 교체 — 격자 4판 실측에서 **8step·캐시ON 이 최적**이었다: 6step·캐시OFF 대비 **80초 빠르고**(521~525s vs 601s) 오디오·화질이 같거나 낫다. ★6step·캐시ON 은 **오디오가 갈라진다**(CEO 육안 \"목소리 갈라지네\") — 속도만 보고 6step 으로 내리지 말 것."
  },
  "m1": {
    "class_type": "MiniMaxH3MemoryEfficientSageAttentionPatch",
    "inputs": {
      "model": [
        "m0",
        0
      ]
    }
  },
  "2": {
    "class_type": "CLIPLoader",
    "inputs": {
      "clip_name": "qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors",
      "type": "minimax",
      "device": "default"
    }
  },
  "3": {
    "class_type": "VAELoader",
    "inputs": {
      "vae_name": "minimax_h3_video_vae_fp16.safetensors"
    }
  },
  "4": {
    "class_type": "VAELoader",
    "inputs": {
      "vae_name": "minimax_h3_audio_vae_fp32.safetensors"
    }
  },
  "101": {
    "class_type": "LoadImage",
    "inputs": {
      "image": "dockaiq-input-47605-input_image-jjin-hero-crop.png"
    }
  },
  "6": {
    "class_type": "MiniMaxH3ReferenceToVideo",
    "inputs": {
      "prompt": "Animate <Picture 1> as one natural 15-second private girlfriend conversation.\n\nVISIBLE REFERENCE:\nContext: brick wall with framed black-and-white photo; large windows showing greenery outside; wooden table and chair nearby\nExpression: smiling softly while covering mouth with hand\nPose: sitting at a table, leaning slightly forward with arms crossed over chest\nSupported gesture: hand partially covers mouth\n\nLOCK THE IMAGE:\nKeep the same identity, face, hair, clothes, pose category, subject size, crop,\nfocal length, visible body coverage, objects, lighting, and background. Start from\nthe uploaded frame and preserve its composition. No new location or unseen body.\n\nPERFORMANCE:\nShe talks affectionately to her boyfriend immediately behind the camera. Emotional\narc: quiet pride with playful knowing eye contact. Use natural eye contact, irregular blinking, breathing,\ntiny listening reactions, restrained smiles, and at most one small gesture already\nsupported by the pose. Keep one continuous physical performance.\n\nDIALOGUE:\nSpeak these four lines in order as one flowing thought:\n<d>[Korean] 네가 좋아하는 걸로 시켰어.</d>\n<d>[Korean] 당연히 기억하고 있지.</d>\n<d>[Korean] 왜 그렇게 놀라?</d>\n<d>[Korean] 나 너 꽤 잘 알아.</d>\nUse natural breath spacing, one brief hesitation, and a tiny laugh only if it fits.\nDo not repeat, rewrite, add, sing, chant, or narrate the words. Match visible mouth\nmotion to each spoken phrase, then let the mouth rest naturally between phrases.\n\nVOICE:\nOne warm young-adult feminine Korean voice, intimate and spontaneous,\nnever robotic or announcer-like. Dialogue and faint room tone only. No male reply,\nno music, no lyrics, no singing, no subtitles, no captions, and no text.\n\nCAMERA:\nOne uninterrupted locked-composition take. No cut, transition, zoom, crop, dolly,\norbit, reframe, focus jump, or scene reset. Only imperceptible stabilization drift.\n\n15 seconds, vertical 9:16, H.264 MP4.",
      "width": 768,
      "height": 1344,
      "length": 124,
      "ref_image_size": "max",
      "clip": [
        "2",
        0
      ],
      "vae": [
        "3",
        0
      ],
      "audio_vae": [
        "4",
        0
      ],
      "ref_images.ref_image_0": [
        "101",
        0
      ]
    }
  },
  "7": {
    "class_type": "BasicGuider",
    "inputs": {
      "model": [
        "m4",
        0
      ],
      "conditioning": [
        "6",
        0
      ]
    }
  },
  "8": {
    "class_type": "MiniMaxH3TurboSampler",
    "inputs": {}
  },
  "9": {
    "class_type": "BasicScheduler",
    "inputs": {
      "scheduler": "simple",
      "steps": 8,
      "denoise": 1.0,
      "model": [
        "m4",
        0
      ]
    }
  },
  "10": {
    "class_type": "RandomNoise",
    "inputs": {
      "noise_seed": 3944676729808669336
    }
  },
  "11": {
    "class_type": "SamplerCustomAdvanced",
    "inputs": {
      "noise": [
        "10",
        0
      ],
      "guider": [
        "7",
        0
      ],
      "sampler": [
        "8",
        0
      ],
      "sigmas": [
        "9",
        0
      ],
      "latent_image": [
        "6",
        1
      ]
    }
  },
  "12": {
    "class_type": "VAEDecode",
    "inputs": {
      "samples": [
        "11",
        0
      ],
      "vae": [
        "vb",
        0
      ]
    }
  },
  "13": {
    "class_type": "VAEDecodeAudio",
    "inputs": {
      "samples": [
        "11",
        0
      ],
      "vae": [
        "4",
        0
      ]
    }
  },
  "14": {
    "class_type": "CreateVideo",
    "inputs": {
      "fps": 24,
      "bit_depth": 8,
      "images": [
        "12",
        0
      ],
      "audio": [
        "13",
        0
      ]
    }
  },
  "15": {
    "class_type": "SaveVideo",
    "inputs": {
      "filename_prefix": "h3_r2v_dockaiq-ref2va_j047605",
      "format": "auto",
      "video": [
        "14",
        0
      ],
      "codec": "auto"
    }
  },
  "m4": {
    "class_type": "H3FirstBlockCache",
    "inputs": {
      "threshold": 0.23,
      "start_step": 2,
      "end_dense_steps": 1,
      "max_consecutive_skips": 2,
      "model": [
        "m1",
        0
      ]
    }
  },
  "vb": {
    "class_type": "H3VideoVAEBatchDecode",
    "inputs": {
      "vae": [
        "3",
        0
      ],
      "enabled": true
    }
  }
}