GALIAIS-Nodes 01 角色身份

The GALIAIS-Nodes character system splits a prompt into ten layers, numbered 01 to 10 in the display names, and 01 角色身份 (Character Identity) is where the character actually becomes a person. This is the node that holds the "who" - subject count, name, source work, artist, age - before the other layers add the face, clothes, and pose. It's the anchor section the Final Composer builds on, so getting identity right here is what keeps a character recognizable across a whole session.

What you fill in

Six content fields, and each has a matching 使用 (use) toggle so you can switch it off without clearing the text:

• 主体人数 (subject count) - 1girl, 2boys, that style of count tag.
• 角色 (character) - the character's name or defining tags. This is the core of "who."
• 作品 (work) - the series/franchise, if any.
• 画师 (artist) - and this one gets special treatment: the node normalizes the artist tag against the Danbooru DB (when one is connected) and inserts it before the subject, because artist tags conventionally lead a prompt.
• 年龄身份 (age/identity) - teenager, adult, loli, occupation, whatever defines the role.
• 身份补充 (identity supplement) - free-text extras, the only multiline field.

Above those sit 角色槽位 (character slot: 全局 / 角色1–8) and 角色标签 (character tag/label). Set the slot to a specific character for multi-character scenes - the composer then groups everything tagged 角色1 under one character and joins characters with your chosen separator. Set it to 全局 and the section applies scene-wide.

Then the common pack machinery: 启用 (enable) and 权重 (weight 0–5) control whether and how hard the section lands, and below that is the random/AI block - 启用随机Tag (enable random tags) flips on automatic tag filling from the connected DB (the pack's Danbooru database), with Tag生成模式 choosing between plain rule-based random, AI-collaborative selection, and AI intent-directed selection. The AI modes need the AI服务商 optional input and cost API calls; the rule modes are local and free. 随机种子 keeps results reproducible.

Outputs and how to read it

Three outputs: 角色段 (the character section object - wire this into the composer's 提示词段 inputs), 文本 (the plain rendered text, useful for previewing), and 元信息JSON (metadata: which fields contributed, what the random picker chose, whether AI wrote anything). When a render ignores your character, 元信息JSON is where you find out that the section was disabled or the 权重 got zeroed.

The honest read

This is the one character layer that's not optional in practice - every other layer describes how a person looks and acts, but identity is the person. The random-tag machinery is the pack's calling card, but for identity you probably want it off: a randomized 角色 field produces a different person every run, which defeats the point of a named character. Save the randomness for outfit and scene. Also: the pack is Chinese-first (all these labels), new, and has no community writeups yet - the README's workflow diagram is the map.

Install

Pack-wide, once:

git clone https://github.com/GALIAIS/GALIAIS-Nodes.git ComfyUI/custom_nodes/GALIAIS-Nodes

Restart ComfyUI, or ComfyUI Manager → "GALIAIS-Nodes". No pip dependencies; the Danbooru DB is a separate download if you want dictionary-backed tag picking.