LoRA Database Stats v03

If you've been using this pack's LoRA nodes for a while, you've got a database full of annotations - but do you know what's actually in it? LoRA Database Stats v03 is the dashboard node: it reads the whole store and tells you how many LoRAs you have, what architectures and categories they break down into, which ones are missing metadata, and what your ratings say about the collection. It's the node you run when you want a status report on your library before deciding what to prune, what to re-test, or what to annotate next.

All four inputs are booleans, which tells you this node is all about report depth rather than configuration. refresh_stats forces a recompute instead of serving cached numbers, show_detailed_breakdown (on by default) adds category and architecture breakdowns, show_missing_data (on by default) lists LoRAs that lack metadata, and show_ratings_analysis (on by default) summarizes the feedback/rating data from the family's Quick Feedback node. Nothing else to set - this is a read-only node, which makes it the safe one in the family to experiment with.

It returns four string reports: stats_overview, detailed_breakdown, missing_data_report, and recommendations. That last output is the interesting one - the node looks at your data and suggests next steps (things like "these 37 LoRAs have no trigger words" or "most of your collection is uncategorized"). It's not AI, just a rules-based read of the numbers, but it's genuinely useful as a to-do list.

Mechanically it's a read pass over the pack's lora_tester_db.json. Nothing moves, nothing writes, nothing gets destroyed - you can run this one a hundred times without risk. That makes it the right node to wire into a workflow you queue periodically, or just to run when you're between projects and wondering what you've accumulated.

The honest take: this node (like the whole family) is single-author, niche, and the "recommendations" are simple heuristics, not wisdom. But a simple heuristic that tells you "you have 214 SDXL LoRAs and 11 are categorized" is already more insight than staring at a folder of filenames gives you. It pairs naturally with the family's maintenance node - stats shows you the gaps, maintenance cleans them up, Batch Info Setter fills them in.

Install the pack and restart, then find it under loaders/lora tester:

cd ComfyUI/custom_nodes
git clone https://github.com/EricRollei/AAA_Metadata_System
cd AAA_Metadata_System
pip install -r requirements.txt

Run it with defaults first. The missing_data_report output is where the actionable stuff lives - that's your annotation backlog, in a single string.