GALIAIS-Nodes Final Composer

Every other node in this pack builds pieces. This one assembles them. The Final Composer (display name: "GALIAIS-Nodes Final Composer") is the hub of the whole 01–10 character workflow: you feed it the section objects coming out of the identity, face/hair/eyes, body, outfit, pose, scene, narrative, object, meta, and NSFW nodes, and it merges them into one clean positive prompt, builds the negative, and hands you the pair ready for a CLIP Text Encode / positive conditioning. It's the pack's pay-off node, and the one you should understand before building anything on top of the character layers.

The inputs that matter

The 16 提示词段 (prompt section) inputs are the workhorses - 提示词段1 through 提示词段16 all accept the GALIAIS_NODES_CHARACTER_SECTION type that the character nodes emit. Wire your character nodes' <section>段 outputs into them in any order; the composer orders sections internally by type, so you don't have to be precious about which numbered input gets which section. The real knobs are the required dropdowns:

• 质量预设 (quality preset) - Anima score_9 / score_8 / score_7, 通用高质量 (generic high quality), or 无. These are literal tag strings: masterpiece, best quality, score_9, safe. Pick the one matching your base model's quality vocabulary - the Anima presets are for Anima, 通用高质量 for the classic anime-SDXL masterpiece, best quality style.
• 负面预设 (negative preset) - same presets the AI Negative Prompt Builder uses: 标准, 轻量, 手部修复, 写实, or 无. This is the base negative the composer starts from.
• 去重 (dedupe, default on) - strips exact duplicate tags across sections. Keep it on; your sections will repeat 1girl without it.
• 多角色模式 (multi-character mode, 自动/开启/关闭) - this is the interesting one. In auto, if any section is tagged to a specific character slot (角色1–角色8 via 角色槽位), the composer groups all of that character's sections together and joins characters with a separator instead of dumping everything into one undifferentiated list. Global sections (slot = 全局) go at the end.
• 角色分隔 (character separator) - 分号 (semicolon, default), 逗号 (comma), or 换行 (newline). How characters are joined in multi-character mode.
• 额外前缀 / 额外后缀 (extra prefix/suffix) - free text before and after everything.

The optional 额外负面 (extra negative) appends to the base negative. Outputs are 正面提示词, 负面提示词, and 提示词JSON (metadata - the section breakdown, quality tags, template used, character grouping). That JSON output is quietly the best debugging tool in the pack: when your render is wrong, it tells you exactly which section contributed what.

Templates

There's also a template layer: 自定义正面模板 (custom positive template) with {{quality}}, {{prefix}}, {{all_sections}}, {{suffix}}-style slots, plus 模板名称 and 模板JSON to pull saved templates from the pack's Composer Template Manager. The template manager article covers that side; here it's enough to know the composer's default behavior already works well without one.

The honest read

This is the one node in the pack worth reaching for even if you ignore the AI stuff. It does the boring-but-vital job - ordered sections, dedupe, quality/negative presets, sane multi-character layout - with zero API calls and zero dependencies. The learning curve is the Chinese labels and the sheer number of inputs, but the defaults are sane: wire sections in, pick a quality preset, run. For most users that's 90% of the value.

Install

Pack-wide:

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

Restart ComfyUI, or search "GALIAIS-Nodes" in ComfyUI Manager. No pip dependencies, no models. New pack, no community writeups yet - but the composer's 提示词JSON output and the README's reference workflow will tell you what you need.