LoRA Stat Viewer

A LoRA is a folder of small matrices pretending to be a file. LoRA Stat Viewer is the window into that folder: feed it any LoRA that this pack has loaded (the LORA output from Load LoRA Only) and it spits out a plain-text report of everything inside - key count, layer types, and the actual key names.

Why would you care? Because the sibling node, LoRA Layers Operation, lets you zero out or scale specific layers - but only if you know what the layers are called. LoRA files from different trainers use different naming schemes, and the default pattern in Layers Operation only knows how to find transformer_blocks.N style keys, which is an SDXL / Illustrious / Flux-era naming. If your LoRA was trained with a different convention, the pattern simply won't match anything and your edit will silently do nothing. This node is the fix: it tells you the exact key names your file uses so you can write a pattern that matches.

The report is one big string on the lora_info output, and it covers four things:

• The total number of keys, so you can sanity-check rank and structure at a glance.
• A grouping by layer type (things like lora_down.weight vs lora_up.weight), showing how many keys of each kind the file carries.
• The first 10 keys, with a full key list appended at the end - this is the part you actually need for writing a regex.
• If it spots any transformer_blocks keys, it lists the block indices present, which are exactly the numbers you'd feed into Layers Operation's layer_indices field.

Mechanically it's as simple as it sounds: the node walks the state dict, groups keys, and joins everything into one readable string. There's no model loading and no magic - which is the right call, because this is a node you run while you're building a workflow, not one that lives in your final graph.

One practical note: the full key list makes this a long string, so don't wire it into something that truncates. Give it a scrollable text display - in ComfyUI that's the core Show Text node, or anything similar from the usual UI packs - and read the output from the node's own panel.

Installing the pack takes about two minutes and needs nothing beyond stock ComfyUI - no extra Python packages, no model downloads. Either use ComfyUI Manager and search for Comfyui-LoraUtils, or:

cd ComfyUI/custom_nodes
git clone https://github.com/lrzjason/Comfyui-LoraUtils

Then restart ComfyUI; the nodes appear under LoraUtils. It's a small pack from lrzjason (aka xiaozhijason on Civitai) with a modest footprint, so be ready to read the source if something behaves unexpectedly.

The honest review: this node is a diagnostic tool, not a day-to-day one. You reach for it when a Layers Operation edit isn't taking effect, and it solves that puzzle in one run. After that it sits in your library until the next weird LoRA rolls through - which, if you download from Civitai with any regularity, will be soon.