Get Index Info

Semantic indexes are opaque blobs - a FAISS file and a SQLite database that do their job silently until something's wrong. Get Index Info is the node that opens the hood. Feed it any SEMANTIC_INDEX and it returns a single STRING containing the index's full health report: name, type, image and vector counts, embedding dimension, which folders it's tracking, storage sizes (FAISS / database / thumbnails), which model built it, when it was created and last indexed - and crucially, whether it has dead vectors that need compacting.

For a beginner this is the "is my library actually healthy?" node. The report reads something like:

Index: portraits
Type: FLAT
Images: 12847
Vectors: 12847
Embedding dim: 4096
Trained: True
Folders tracked: 3
Storage:
  FAISS: 52 MB
  Database: 9 MB
  Thumbnails: 148 MB (12847 files)
Model: Qwen3-VL-Embedding-8B
Last indexed: 2026-08-10

The one line to watch is the compaction block. If you've removed folders or files, deleted_vectors climbs, wasted_percentage shows how much of the FAISS file is dead weight, and the report literally tells you when compaction is recommended - the pack's own threshold is ~10% waste. See Compact Index for the fix.

Inputs, output, and how to use it

One required input (index), no options, one info STRING output. Wire it into a Show Text node (or any text display) to read it. It's a debug/ops node, so it sits in the maintenance chain - typically after Load/Create Index, alongside Remove Folder and Compact. It doesn't change anything; it only reports.

Install

Same pack install as the rest - ComfyUI Manager (search "Semantic-Search", pack "Eric") or clone https://github.com/EricRollei/Semantic-Search into custom_nodes, install the requirements.txt deps (faiss-cpu, qwen-vl-utils, Pillow, numpy), restart, and look under Eric/SemanticSearch.

Where people get burned

• You see stats but no folders listed. Folders tracked only lists folders if the index was built by Add Folder to Index. An index that was created empty - or whose metadata got out of sync - will report honestly, and "Folders tracked: 0" is your cue that something's off with how it was populated.
• Vectors vs Images disagree. A mismatch means deleted entries left orphans behind (vectors > images), or a folder was removed without the vector cleanup running. That's normal-ish, and the report flags it via the compaction block. If vectors and images look right but search returns nonsense, check Embedding dim against your model.
• The path storage question again. The info output doesn't tell you where the index lives; that's INDEXES_PATH in core/config.py (default H:/semantic_search/indexes). If you move machines, repoint that path or your indexes "disappear" while Load/Create Index happily makes fresh empty ones.

It's the most boring node in the pack and also the most reassuring. Five minutes with Get Index Info before you blame your search results for being wrong will save you from chasing phantom accuracy problems that were really just an index that needed compacting.