GALIAIS-Nodes Prompt Template

The Prompt Template is the node that ties this pack's whole slot philosophy together. You define a prompt skeleton with {{placeholder}} holes in it, fill the holes with dedicated inputs, and it renders the finished positive and negative prompts. It's the boring-but-powerful middle of the GALIAIS flow: one template, many fillings.

The two ways to get a template

模板来源 switches between "手写" (handwritten) and "词典模板ID" (dictionary template ID).

• Handwritten is what you'll start with. Type your skeleton into 正面模板 (positive) and 负面模板 (negative), using {{质量}}, {{主体}}, {{角色}} and friends as the holes. The 13 optional slot inputs - quality, subject, character, appearance, outfit, pose, expression, camera, scene, lighting, style, detail, natural language - are what get substituted in. Each one is forceInput, so you can wire them from other nodes instead of typing.
• Dictionary template ID pulls a stored template out of the Danbooru DB by ID or name (the DB's prompt_templates table). You need a real DB path for this mode, and the downloaded template's positive/negative text becomes the skeleton. Everything else works the same.

The slot substitution is pleasantly forgiving: {{quality.style}} falls back to the plain {{风格}} slot if the dotted path isn't filled, so a slightly-misnamed placeholder degrades gracefully instead of crashing or spitting a literal {{...}} at the sampler.

What comes out

正面提示词 and 负面提示词 are your strings, plus 元信息JSON records which template and slots were used. One genuine gotcha: the positive output runs through the pack's Anima-flavored joiner, which strips quality tags (masterpiece, best quality, score_9) and any @artist token by default. If you're targeting a Pony or Illustrious checkpoint that wants score_9 in the positive, this node will quietly delete it. Plan your template accordingly - put quality tags in the negative side or accept the Anima cleaning.

Where it fits

Templates are how you keep a consistent structure across a whole batch. Lock the skeleton, vary the slots (feed 场景 from a random pool or a DB selector), and every generation shares the same anatomy while the details rotate. It's the reusable layer the one-shot Typed Composer V2 doesn't give you.

Installing

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

Restart ComfyUI. No dependencies. Dictionary-template mode needs the Danbooru runtime DB loaded (build it with Danbooru Runtime DB Builder, load it with the DB Loader), but handwritten mode runs with no DB at all. And yes - the slot inputs are labeled in Chinese (主体 is "subject", 场景 is "scene"). The tooltips help, but you'll memorize them in one session.