LoRA Info Inspector 🔬

Every LoRA file on your disk is a claim: "this small adapter will push the checkpoint in this direction." LoRA Info Inspector is the auditor that checks the claim before you believe it. It reads a LoRA file and reports its structural metadata - key count, rank, alpha, and which modules it targets - without loading it into a pipeline and without touching your model. You get a facts sheet on the adapter, not a patched model. It's the node you run when you want to know what you're about to stack, before you find out the hard way.

How it works

It inspects the file's structure and hands you four things on wires:

• key_count - how many tensors the file contains. A rough sense of size and complexity.
• rank - the rank of the down-projection. The classic LoRA dial: lower rank = more compact but less expressive; higher = more capacity and more VRAM and overfitting risk. Combined with alpha it tells you how the adapter was built.
• alpha - the scaling constant baked into the file. The ratio of alpha to rank is what actually controls effective strength, which is why two "1.0" LoRAs can hit very differently.
• target_modules - which parts of the model the LoRA patches (attention projections, for instance). This is the compatibility check: a LoRA targeting SDXL attention layers is not going to do anything useful on a Flux model, and seeing the target list is how you know before wasting a render.

Plus summary, a text overview of the whole audit.

Where it fits

Three honest uses:

1. Auditing before use. Pull a LoRA from the internet and you don't know its rank or what it touches. One pass with this node and you do - before it's mixed into a workflow and doing something unexpected.
2. Debugging a "why is this LoRA so weak/so strong?" problem. The alpha-to-rank ratio explains a lot of surprising strength behavior that the weight dial alone can't.
3. Building debug pipelines. Because it's a non-destructive inspector with typed outputs, you can gate or branch on what it finds (e.g. route differently based on rank) without loading the file.

The one thing it is not: a substitute for the pack's LyCORIS Format Inspector when format matters. That node answers "is this actually a LoRA, or a LoHa/LoKr/DoRA?" - this one answers "what are its dimensions and targets." Run the format inspector first if you're unsure what family the file is; run this one if you want its vitals.

Install

No extra dependencies:

cd ComfyUI/custom_nodes
git clone https://github.com/TensorVizion/OmniNodes

Restart, find it under TensorVizion/Model Utilities (ComfyUI Manager, search "OmniNodes", also works).

Troubleshooting

• "LoRA not found" - the file must be in ComfyUI's registered loras folder (ComfyUI/models/loras/), not an arbitrary path. That's the pack's standard, documented gotcha.
• rank reads as 0 or odd values - some files (especially non-LoRA formats) don't expose rank the way a plain LoRA does. If you see a nonsense rank, run LyCORIS Format Inspector - you may be looking at a LoHa or LoKr that doesn't structure its rank the same way.
• target_modules looks empty - could be an unusual file layout; cross-check with the summary text.

There are heavier LoRA management suites out there that manage entire libraries. This node isn't that. It's a focused, typed, wireable inspector for a single file - which, for auditing one adapter before you commit a render to it, is exactly the right amount of tool.