LoRA Info Loader

The CheckpointInfoLoader idea, applied to LoRAs: a LoRA you downloaded usually came with example images, and those PNGs frequently carry an embedded ComfyUI workflow and generation metadata - trigger words, recommended strength, the sampler settings the author actually used to show off what the LoRA does. Normally getting at that means dragging the preview image itself onto your canvas. This node does the lookup inside a workflow instead: point it at an installed LoRA and pull the preview, the embedded workflow, and the metadata as outputs.

What it does

lora picks the source file from a dropdown of your installed LoRAs. workflow_index (default 0, range 0–100) selects which associated example to pull if the LoRA has more than one. Outputs: preview (IMAGE), workflow_json (the embedded workflow graph as a string), and metadata_json (whatever other generation info travels with that example - string).

Why you'd actually want this

LoRAs are the one model type where "how was this actually meant to be used" is a genuinely common question - recommended strength varies a lot between LoRAs (the old SDXL-era 0.5–0.8 rule of thumb doesn't hold on newer architectures, where some LoRAs want 1.0 or more or they lose likeness entirely), and trigger words are essential and easy to lose track of once you've downloaded a folder of a few dozen LoRAs with generic filenames. If the author's example image has that information embedded - either as workflow metadata or a caption baked into the PNG's generation params - this node surfaces it without you needing to go find the original download page again.

It pairs naturally with LoRAMetaKeys in this pack: LoRAMetaKeys tells you the internal structure (tensor names, training metadata if the file's header carries it), while this node tells you the intended usage (example workflow, recommended settings) - two different questions, both worth asking about a LoRA you're not familiar with before you build a workflow around it.

Installing it

ComfyUI Manager: search Model Utility Toolkit, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/silveroxides/ComfyUI-ModelUtils

Restart ComfyUI. No downloads needed for the node - what it finds depends on whether the LoRA has a preview image with embedded data at all.

Where people get burned

Same honest limitation as the checkpoint version: this reads what's already there, it doesn't fetch anything. A LoRA with no preview image, or one where the uploader never embedded workflow metadata into the example, gives you empty workflow_json/metadata_json outputs - that's the file, not a bug in the node. Plenty of LoRAs on CivitAI and elsewhere ship with example images that are just images, no embedded generation data at all.

workflow_index only matters if a LoRA actually has multiple stored examples to pick between; if bumping it doesn't change the output, that LoRA likely only has one example attached. And treat any recommended strength you pull from a workflow as a starting point, not gospel - it reflects one specific base checkpoint and sampler setup the author happened to use, and your own workflow's base model, sampler, and other stacked LoRAs can shift what strength actually looks right.