GALIAIS-Nodes 09 Meta技术

The 09 Meta技术 (Meta & Technical) layer is the pack's junk drawer with a purpose: quality-control tags, technical render notes, and process flags that don't belong in any visual layer. In the pack's ten-layer scheme it's the last "normal" section before the NSFW layer, and it's where you tell the model about render quality and where you tell yourself about outstanding issues.

What you fill in

Four content fields plus a supplement, each with a 使用 toggle:

• 管理质量 (quality management) - the quality-tag block: masterpiece, best quality, highres, detailed and friends. This is where your base model's quality vocabulary lives.
• 技术规格 (technical specs) - render-level tags: highres, detailed shading, best quality, resolution/quality specifiers.
• 覆盖处理 (coverage/overlay handling) - tags about how elements overlap or are composited: layered, transparent background, drawn on top style notes.
• 待复审 (pending review) - this is the honest-to-goodness process field: a place to park a note or tag for things you need to revisit (the author's own "to review" flag), rather than trying to render a reminder into the image.
• Meta补充 (Meta supplement) - free-text extras, multiline.

Plus the shared layer controls: 角色槽位 (全局 / 角色1–8), 角色标签, 启用, 权重, and the random/AI block (启用随机Tag, Tag生成模式, 随机种子, optional AI服务商), with optional DB, 上游上下文, and 上游提示词段.

Outputs

The standard trio: Meta段 (section object → composer), 文本 (plain text), 元信息JSON (contribution + random audit).

Where it earns its keep

The quality-tag fields are the point for most users: putting masterpiece, best quality in a dedicated field rather than the top of a random prompt line keeps quality tags grouped and weight-controlled, which is exactly how the pack's Final Composer expects them. And 待复审 is a quietly nice idea - a persistent scratch field that travels with the workflow JSON instead of living in your head. You'll find yourself using it on day two.

The honest read

This layer overlaps heavily with the Final Composer's 质量预设 (quality preset) dropdown, so the honest advice is: pick one place for quality tags. If you're using the composer's preset, leave 管理质量 and 技术规格 empty here and avoid double-stacking masterpiece, best quality - dedupe catches exact repeats but the tags come from two sources and read as one list anyway. If you're not using a preset, this is the right home for them. Also note: on 2026 flow-matching models with LLM encoders, the old quality-tag stack does less than it did on SDXL anime models, so match this field's vocabulary to your base model.

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 separate if you want dictionary-backed picking. New pack, no community writeups - lean on 元信息JSON and the README.