Eric Krea2 Diagnose LoRA

LoRAs are the black boxes of this hobby. You download a file, apply it, and the result either looks right or it doesn't - and when it doesn't, there's usually no way to tell whether the file is broken, misformatted, or just trained wrong. Eric Krea2 Diagnose LoRA is the pack's answer: it inspects a LoRA without applying it and hands you a text report about what it found.

The key word is "without applying." This is a read-only node - it's the diagnostic sibling to Apply LoRA and the Multi-LoRA Stack, and it exists because the author (who A/Bs fine tunes for a living, judging by this pack's changelog) got tired of debugging misbehaving adapters blind. Wire it up wherever you'd wire Apply LoRA and it inspects the same file, then reports instead of modifying the pipeline.

Inputs are minimal. pipeline (required, the KREA2_PIPELINE) - it needs the pipeline context to interpret what it's looking at. lora_name is the dropdown over models/loras/Krea2/, and lora_path_override covers files elsewhere. One output: report, a STRING. The node is marked as an output node, so you can view the report directly; wire it to a text display if you want it in the UI.

What does the report actually contain? The README positions it alongside the Apply node's trigger handling, so the practical job list is: the LoRA's metadata (trigger words if any are embedded in the safetensors), the target module set, and whether the file structure matches what the pack's loader expects. In other words - the things that make a LoRA either work or silently do nothing. When a downloaded Krea 2 LoRA "doesn't work" and the actual cause is a wrong-format file (FAL-format LoRAs need conversion before ComfyUI will load them - a known Krea 2 ecosystem gotcha) or a trigger word you were never told about, this node is where you find out in seconds instead of after a dozen wasted generations.

When would you actually reach for it? Three honest cases: a new LoRA behaves like nothing happened, so you check whether it loaded and what it contains; you're about to build a Multi-LoRA stack and want to confirm the files aren't duplicating each other's targets; or you're deciding whether to tune a LoRA's strength or drop it entirely - the report tells you what the file is claiming to do before you spend generations figuring out it doesn't.

It won't tell you whether a LoRA looks good. That's the eye test. But "is this file the right shape and does it carry the metadata it should" is exactly the kind of question you shouldn't have to answer by trial and error.

How to install

Part of EricRollei/Krea2_ComfyUI_Advanced. ComfyUI Manager → search "Krea2_ComfyUI_Advanced", or:

cd ComfyUI/custom_nodes
git clone https://github.com/EricRollei/Krea2_ComfyUI_Advanced

Restart. Pack-wide requirements apply (git diffusers with Krea2Pipeline, transformers>=4.57.0, Krea 2 weights for the pipeline input).

Common issues

• LoRA doesn't appear in the dropdown - models/loras/Krea2/ only; anything else needs lora_path_override.
• Report says the file's wrong shape / won't load - for Krea 2 specifically, FAL-format LoRAs must be converted before ComfyUI will touch them. This is the classic "I downloaded a LoRA and it does nothing" root cause, and it's exactly what this node exists to surface.
• Nothing to diagnose - it's read-only; you still need Apply LoRA or the Stack to actually use the adapter.