GALIAIS-Nodes Image Detail Blueprint

The category says ai, the name says "blueprint," and here's the pleasant surprise: this node calls no API and needs no key. Image Detail Blueprint is a deterministic prompt-analysis engine. You feed it a Danbooru-style tag prompt, and it prunes conflicts, builds a structured generation plan, and renders natural-language drafts - all locally, all reproducible, no LLM in the loop. That's also why the pack trusts it as the first stage of its one-click Prompt Orchestrator.

Think of it as a translator between two dialects: the tag soup your resolver produced, and the natural-language description that some models (and your own sanity) appreciate. It won't invent flourishes the way an LLM would - which is the point. You get consistency instead of surprise.

How it works

You give it a 正面提示词 (positive prompt) plus three settings: 输出语言 (Chinese / English / mixed), 细节强度 (refined / standard / detailed), and 允许NSFW. Then there's the interesting one: 启用冲突剔除 (conflict pruning, on by default) with a 冲突剔除策略 (automatic / keep-first / keep-last). Conflict pruning is where deterministic logic does the heavy lifting - it looks for contradictory tags and removes them before anything else happens, so your prompt doesn't say both "long hair" and "short hair."

From the pruned prompt it builds a "generation plan" and three blueprints: a caption blueprint, a scene-design blueprint, and a full-image-detail blueprint. The DB (optional, wired from the Danbooru DB Loader) supplies tag context; without it, the node still works, it's just less context-aware.

Four outputs:

• 完整蓝图JSON - the whole structured plan, machine-readable.
• 场景导演描述 (scene director description) - natural language describing the scene; the one you'd read or paste into a natural-language model.
• 完整自然语言草稿 (full natural-language draft) - the complete prose draft of the image.
• 诊断JSON - diagnostics on the input prompt (what resolved, what didn't), useful for spotting junk before you render.

For a beginner: 场景导演描述 is the human-facing output, and 诊断JSON is your debugging tool.

Installing it

It ships in 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, no API setup - the absence of an AI provider input is the whole story here. Field labels are Chinese, as throughout the pack.

Why it earns its place

Two workflows love this node. First, as a drafting stage: let the blueprint produce natural language from your tags, and feed that prose to an Anima- or Z-Image-style model that wants sentences, not tag lists - the KB's core lesson that tag-and-prose models coexist in one prompt. Second, as the deterministic first pass before an AI enricher: giving an LLM a cleaned, conflict-free blueprint to work from causes far less subject drift than letting it rewrite your raw tag soup. The pack's own orchestrator does exactly this. The one trap is expecting LLM-like creativity from it - it's a structured generator, not a poet, and that's a feature.