GALIAIS-Nodes Prompt Quality Score

Sometimes you don't need a yes/no verdict - you need a number you can plot. The Prompt Quality Score is the lighter sibling of the pack's Quality Gate: same scoring engine, no threshold, no pass/fail drama. Feed it a prompt, get back an integer score, a letter grade, and a JSON blob explaining exactly where the points went.

It shares the pack's diagnostic routine, so it checks the same things the gate does: duplicate tags, conflicts between known groups (smile vs crying, open_eyes vs closed_eyes, simple_background vs detailed_background), unknown tags, NSFW tags sneaking into a SFW prompt, and whether your prompt covers the basics - subject, clothing, pose, scene, style. The score starts at 100, sheds points per problem, adds a completeness bonus, and clamps to 0–100. Grades fall out of the score: A at 90+, B at 75+, C at 60+, D below.

The inputs

• 提示词 - the prompt being judged (labeled in Chinese; that's "prompt").
• 允许NSFW - off by default. Turn it on when you're genuinely prompting NSFW, or the scorer counts every NSFW tag as an error and the grade collapses.
• 项目配置 and DB (optional) - a project config can carry allow_nsfw and a DB path; the DB is what lets the scorer actually resolve tags against the Danbooru dictionary.

Where it shines

Because the score is just an INT, it's the node to reach for when you want to tune rather than gate. Wire 质量评分 into a scheduler or an XYZ-plot-ish loop and sweep a section of the prompt while watching the number move. It's also a handy QA stop when you're about to batch a hundred variations - a quick score tells you if a section edit nuked completeness before you render the batch.

The 评分JSON output is the real prize. It's a structured report: duplicates, unknown, nsfw_tags, conflicts, order_warnings, and a per-dimension completeness map. If you're building automation on top of this pack - say, a workflow that auto-rewrites weak sections - that JSON is what you'd parse.

Installing and the DB caveat

Same as every node in this pack:

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

Restart ComfyUI. Zero pip dependencies. But note: without a Danbooru runtime DB loaded (via the DB Loader), the scorer can't distinguish known from unknown tags and can't detect NSFW tags at all - so unknown is empty, NSFW checks are blind, and the number drifts optimistic. Load the DB if you want the score to mean anything.

One honest warning: this is a heuristic, tuned around Danbooru-tag conventions and the Anima-style ordering this pack targets. It's a fine consistency check for anime SDXL-lineage prompts, not a universal judge of prompt quality - don't treat an A as a guarantee of a good render.