Eric's Metadata Filter V2

"Which of my 4,000 output images used this LoRA?" is a question no file explorer answers. Eric's Metadata Filter V2 does: point it at a folder, tell it which metadata field to look at, and it hands back the matching files, a count, and optionally the images themselves. If you keep a metadata database on your generations, this is the node that turns it into a search engine.

It's part of the AAA Metadata System pack's metadata family, and it's a power tool for anyone who's been saving with metadata-aware nodes. The pack's metadata structure records things like prompt, model, rating, aesthetic and technical scores, LoRA usage, and face/blur analysis - this filter queries all of that. It's also useful for QA: "show me everything in this folder rated below 3" or "everything that used the broken model."

How it works

You give it a folder and a query. It searches either the pack's SQLite database (fast, structured) or falls back to scanning the folder and reading each image's metadata directly. The field_type dropdown picks what you're filtering on, operation picks how, and value is the needle. The search is substring-friendly by default (include_substrings on, case-insensitive unless you say otherwise), and results can be sorted and limited.

The advanced_query field is the escape hatch: when the friendly dropdowns aren't enough, you can drop in a structured query there instead. If the results ever look wrong and use_database is on, try switching it off to scan files directly - the two paths can disagree if the database is stale.

Inputs and outputs that matter

The three that define a query:

• folder_path - where to look (required, string).
• field_type - keywords, aesthetic_score, technical_score, prompt, model, directory, lora, any_text, rating, faces, blur, or custom. custom plus custom_field is how you query something not in the list.
• operation - contains, equals, greater_than, less_than, between, exists, missing. Pair between with value and value2.

The limit (default 10) caps results so you don't get a 4,000-item dump; load_images decides whether the actual images come back as an IMAGE output or just paths.

Outputs: matched_paths (the file paths, as one string), match_count (how many matched), and images (an IMAGE batch, only if you asked for it).

Installing it

Ships in AAA Metadata System by Eric Hiss (GitHub: EricRollei). Install via ComfyUI Manager (search "AAA Metadata System") or:

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

Restart ComfyUI. Nothing extra to download.

Common issues

The most common "failure" is empty results, and it's almost always the same root cause: the images in that folder have no readable metadata because they weren't saved by a metadata-writing node, or the database is empty. The filter is only as good as the data - no metadata, no matches. If you're sure the metadata exists, check field_type: filtering aesthetic_score when the images only have prompts will return nothing even though prompt would. And watch limit - if you expect 50 matches and get 10, that's the cap doing its job, not a bug. For a slightly friendlier version of this node with fewer knobs, the pack also ships Enhanced Metadata Filter V2.