LoRA Batch Info Setter v03

If you have more than a couple dozen LoRAs, you already know the problem: filenames like fancystyle_v2.safetensors tell you nothing, and you can't remember which is SD1.5, which is SDXL, or what any of them are for. LoRA Batch Info Setter v03 is part of this pack's "LoRA Tester" family, and its job is simple: annotate a whole directory of LoRA files with architecture, category, and default strengths in one run, so your library becomes searchable and testable instead of a pile of mystery weights.

The inputs are refreshingly few. dir_path points at your LoRA folder, recursive (on by default) dives into subfolders, and file_pattern defaults to *.safetensors. Then come the values you're stamping onto everything: architecture (SD1.5, SD2.1, SDXL, Pony, Illustrious, Noobai, Flux, and more - 13 choices), category (style, character, concept, pose, clothing, and others, defaulting to unknown), and the two strengths model_strength (0.8) and clip_strength (1.0) that will be proposed for each LoRA. There's an overwrite_existing flag (off by default) so you don't clobber annotations you already made, and a notes field for batch-level context. Single output: status.

Mechanically, it's writing into the pack's LoRA database - a JSON file (lora_tester_db.json) keyed by file hash, the same store the other LoRA nodes in this family read from and write to. So the flow is: annotate with Batch Info Setter, browse with the Gallery nodes, query params with the Params Loader. The "Tester" in the family name is the pack's broader concept of a LoRA-rating workflow - this node is the bulk-annotate step that makes the rest of it usable.

The honest advice: overwrite_existing defaults to false for a reason. If you've already curated categories for your best LoRAs, a batch run will respect that until you explicitly flip the flag. And "Unknown" architecture is not a failure state - the family's Bulk Operations node has an auto-detect pass that can fill it in later, so it's fine to batch-set category now and backfill architecture.

One more thing to know: this is a single-author pack with essentially no community presence, so the database schema is whatever this author decided it should be. If you later switch to a different LoRA manager, your annotations live in this pack's JSON file, not in the .safetensors themselves. Worth knowing before you invest an hour annotating.

Install the pack and restart, then it's 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

Point dir_path at your models/loras folder, pick your architecture, run it. Ten minutes of setup turns a folder of mystery files into a labeled library you can actually search.