LoRA Info Setter v03

Every LoRA you test generates knowledge you immediately forget: the trigger words, the CFG that worked, which checkpoint it pairs with. LoRA Info Setter v03 is the node that writes it all down. It takes a single LoRA file and records a full "library card" for it - architecture, category, trigger words, recommended strengths, and compatibility notes - into the pack's LoRA database. If the batch setter is for stamping a whole folder, this is for giving one LoRA the full treatment, and it's the input side of the family's Params Loader.

The required inputs are the card itself: lora_path (the file), architecture (13 choices from SD1.5 to Flux), category (style, character, concept, pose, and more), notes, and the trigger-word pair trigger_words_full and trigger_words_selected - both comma-separated strings, where full is everything the LoRA knows and selected is the subset you actually use. Then the strengths you found work: model_strength (0.8) and clip_strength (1.0), plus trigger_placement (beginning vs. end of prompt). The optional section captures the compatibility research: compatible_checkpoints (one path per line), compatible_loras, and placement_notes for the "this trigger needs to go at the start or it doesn't work" wisdom you only learn by testing.

Output is a single status string confirming the write. That's it - this is a write-only node. The read side is the family's LoRA Params Loader, which pulls all of this back out later so you can drop it into a prompt or a loader.

Mechanically it writes to lora_tester_db.json, the same store the whole family shares, keyed by file hash. No network, no model loading - pure bookkeeping. The useful detail is that you can fill in just the fields you know and leave the rest; there's no requirement to complete every field before saving, so you can note the trigger words now and add compatibility checkpoints after you've tested it more.

Where it fits: this node is the reason the family works. A LoRA with its trigger words and strengths saved is one you can actually re-run weeks later; a LoRA with nothing saved is a mystery again. It's niche, single-author, zero community - but for anyone testing a lot of LoRAs, an annotation workflow is worth more than any gallery UI.

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

Set lora_path, fill in the trigger words you found, pick architecture and category, run it. Ten seconds now saves you ten minutes of re-discovery later.