{
  "id": "be1e85b1-4ede-4c33-a207-02c277979872",
  "revision": 0,
  "last_node_id": 4,
  "last_link_id": 2,
  "nodes": [
    {
      "id": 1,
      "type": "MiniMaxH3RefModFolderLoader",
      "pos": [
        40,
        160
      ],
      "size": [
        520,
        360
      ],
      "flags": {},
      "order": 0,
      "mode": 0,
      "inputs": [],
      "outputs": [
        {
          "name": "refs",
          "type": "H3_REF_LIST",
          "links": [
            1
          ],
          "slot_index": 0
        },
        {
          "name": "count",
          "type": "INT",
          "links": [],
          "slot_index": 1
        }
      ],
      "properties": {
        "Node name for S&R": "MiniMaxH3RefModFolderLoader"
      },
      "widgets_values": [
        "/workspace/datasets/CHANGE_TO_YOUR_FOLDER",
        8,
        16,
        1280
      ],
      "title": "1 | UPLOAD CHARACTER PHOTOS"
    },
    {
      "id": 2,
      "type": "VAELoader",
      "pos": [
        40,
        600
      ],
      "size": [
        520,
        100
      ],
      "flags": {},
      "order": 1,
      "mode": 0,
      "inputs": [],
      "outputs": [
        {
          "name": "VAE",
          "type": "VAE",
          "links": [
            2
          ],
          "slot_index": 0
        }
      ],
      "properties": {
        "Node name for S&R": "VAELoader"
      },
      "widgets_values": [
        "minimax_h3_video_vae_fp16.safetensors"
      ],
      "title": "H3 video VAE - leave selected"
    },
    {
      "id": 3,
      "type": "MiniMaxH3RefModExtract",
      "pos": [
        680,
        160
      ],
      "size": [
        560,
        850
      ],
      "flags": {},
      "order": 2,
      "mode": 0,
      "inputs": [
        {
          "name": "refs_bundle",
          "type": "H3_REF_LIST",
          "link": 1
        },
        {
          "name": "vae",
          "type": "VAE",
          "link": 2
        }
      ],
      "outputs": [
        {
          "name": "mods",
          "type": "H3_REF_MODS",
          "links": [],
          "slot_index": 0
        },
        {
          "name": "details",
          "type": "STRING",
          "links": [],
          "slot_index": 1
        }
      ],
      "properties": {
        "Node name for S&R": "MiniMaxH3RefModExtract"
      },
      "widgets_values": [
        "my_character_v1",
        "Full Reference",
        "identity",
        0.0,
        768,
        16,
        16,
        16,
        0,
        false,
        false,
        1,
        0,
        "Reference photos of one character.",
        true,
        "manual",
        "",
        "error"
      ],
      "title": "2 | CHANGE NAME, THEN RUN"
    },
    {
      "id": 4,
      "type": "Note",
      "pos": [
        40,
        790
      ],
      "size": [
        560,
        730
      ],
      "flags": {},
      "order": 3,
      "mode": 0,
      "properties": {},
      "widgets_values": [
        "CHARACTER REFMOD CREATOR\n\n1. Choose 5-8 clear photos of ONE character. One image works, but multiple views are useful. Include front, three-quarter, profile and a wider shot. Avoid blur, heavy filters, sunglasses and near-duplicates. Captions are not needed.\n2. Click Upload photos, or drop 1\u20138 selected photos onto the first node's upload area. Wait for photos ready. PNG, JPG, WebP or BMP; up to 20 MB each. The folder and count are set automatically.\n3. Set a unique name such as alex_h3_v1 in the Create node. Existing files with the same name are replaced when saved.\n4. Leave the other settings as supplied and click Run. This VAE-only creation workflow does not load H3's video generation model or train a LoRA.\n5. The saved file is /workspace/models/refmods/<name>.safetensors. In your generation workflow, refresh RefMods and select it in Load H3 RefMods.\n\nFull Reference preserves detail. Refinement Steps, pooling and merge are unused in this mode. concept_type=identity is descriptive metadata; it does not train facial recognition. max_tokens=0 prevents dropping reference views to fit a cap, so more photos increase generation cost.\n\nCROP NOTE: this simple creator stacks multiple photos into ONE video-kind reference, cropping them to the first photo's aspect ratio. Use matching aspect ratios and keep faces well inside the frame. With a single photo it creates an image-kind reference.\n\nPROMPT LABELS: with H3 RefMod Text Encode, a multi-photo file made here normally appears as <Video 1>, not eight separate <Picture> labels. Use the encoder's reference_map output to check. Separate-image bundles use Picture labels for each view. Always update the character-reference lines when replacing files.\n\n\nUPLOAD SETUP: Install the RefMod Pilot repository under ComfyUI/custom_nodes, restart ComfyUI, then refresh the page. This is a ComfyUI extension, not a browser add-on. Uploading does not run the workflow.\n"
      ],
      "title": "READ ME / four simple steps",
      "color": "#432",
      "bgcolor": "#653"
    }
  ],
  "links": [
    [
      1,
      1,
      0,
      3,
      0,
      "H3_REF_LIST"
    ],
    [
      2,
      2,
      0,
      3,
      1,
      "VAE"
    ]
  ],
  "groups": [],
  "config": {},
  "extra": {
    "ds": {
      "scale": 0.62,
      "offset": [
        30,
        20
      ]
    }
  },
  "version": 0.4
}