GALIAIS-Nodes Project Config

GALIAIS-Nodes has a lot of settings, and most of them are the same across a whole project: which dictionary you use, whether NSFW is allowed, what format the prompts come out in, what the default negative is, and - most importantly - the AI endpoint. The Project Config node exists so you set those once and wire its outputs everywhere, instead of re-typing an API key into a dozen nodes.

It's the pack's central configuration object, and its two most-used outputs double as the AI provider and the embedding provider that other nodes accept as inputs.

How it works

The inputs split into project defaults and AI credentials:

• DB路径 (DB path) - your dictionary. The DB optional input lets you take it from an already-loaded DB instead.
• 默认语言 - currently only zh-CN is offered. The whole pack is Chinese-first, and this is the default all its nodes inherit.
• 允许NSFW - the project-wide NSFW policy. Nodes that take a project config (Prompt Inspector V2, Prompt Profile) inherit this instead of asking you again.
• 默认输出格式 - Anima (the pack's default; Anima-style tag+score formatting) or 通用Danbooru (general Danbooru).
• 默认去重 (default dedupe) and 默认负面预设 (default negative preset) - the project-wide defaults.
• AI服务商URL, AI密钥或环境变量, AI模型, Embedding模型 - the AI setup. The key accepts a literal value or env:VAR_NAME / $VAR_NAME to read from the environment. The node splits LLM vs. embedding models automatically based on the name (it recognizes bge-, e5-, jina-embeddings and friends as embedding models).
• Tag黑名单 (optional) - bake a blacklist into the whole project.

Four outputs: 项目配置 (the config object - wire into nodes that accept it), AI服务商 (the ready-to-use LLM provider), Embedding服务商, and 配置JSON (the human-readable dump).

Installing it

It's part of the GALIAIS-Nodes pack:

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

Restart ComfyUI, or install via ComfyUI Manager ("GALIAIS-Nodes"). No Python dependencies. The dictionary is the only external asset you must supply.

The gotcha - and the payoff

The trap is key hygiene: if you paste a literal key here, it flows into the config JSON and any workflow that displays it. Prefer env:MY_KEY - the node resolves it from the environment and the emitted config masks the secret. That's the right habit in a workflow format that gets shared as image metadata.

The payoff is that Project Config turns into the single lever for your whole graph: switch dictionary, flip NSFW, swap AI models, and every node downstream that takes 项目配置 or AI服务商 follows along. If you're building one of this pack's bigger workflows - say, a character pipeline with an orchestrator - this node is where you configure it, not in fifteen individual places.