Lora Loader with info

Loading a LoRA is a solved problem - the core LoraLoader has handled it since 2023, and this node is a straight clone of it. So the only reason to reach for Lora Loader with info is hiding in the name: the info. On top of the normal MODEL and CLIP outputs, it hands you the LoRA's filename and both strength values as plain strings. That sounds trivial until you're trying to record which LoRA produced which image in a batch, or building generation metadata into the workflow itself.

Quick LoRA refresher, because the names matter. A LoRA is a small adapter (10–200 MB) that patches a checkpoint instead of replacing it. The strength_model slider controls how hard it pushes the model's weights; strength_clip controls how much it reshapes the prompt conditioning. Both default to 1.0, which is often too strong - the community standard is closer to 0.5–0.8 for many style LoRAs. It's stackable: a style LoRA plus a character LoRA plus a concept LoRA is normal, though their interactions get unpredictable fast.

How it works

Mechanically it's the same call the core node makes - comfy.sd.load_lora_for_models - with two small additions. It strips the .safetensors off the filename before returning it, and it caches the loaded file per node instance so it doesn't re-read from disk on every run. Nothing exotic, and nothing that should break.

Inputs and outputs that matter

• model and clip - wire these straight from your checkpoint loader, exactly as you would with the core node.
• lora_name - dropdown populated from ComfyUI/models/loras.
• strength_model and strength_clip - the two sliders above, both ranged -20 to 20 with step 0.01.

Outputs are MODEL, CLIP, and then the reason you're here: lora_name, strength_model, and strength_clip as STRINGs. Feed those into the same pack's Image Info node (its lora_list wants a flat list, three entries per LoRA - name, model strength, clip strength - which is exactly this node's three string outputs in order), or into a filename builder so every saved image says "made with my_style at 0.7". That's the ComfyUI metadata habit in miniature: the workflow and its settings travel with the output.

Where the sharp edges are

Two gotchas, one of them a genuine trap.

1. Don't set both strengths to 0. The code has an early return: if strength_model == 0 and strength_clip == 0 it bails out with just (model, clip) - but the node declares five outputs. ComfyUI chokes on the missing three and errors the node. To disable a LoRA, bypass the node instead of zeroing it. If you genuinely want a near-zero effect, use 0.01, not 0.
2. It's one LoRA per node. To stack several you chain multiple of these in series, which is the old-school way. If you routinely run five LoRAs, look at rgthree's Power Lora Loader instead - it stacks in one node, adds per-entry toggles, and reads trigger words off the file. This node does none of that; it doesn't even show trigger words.

Also worth knowing: if you don't need the string outputs at all, the core LoraLoader does the identical job and you can skip this pack entirely. And remember the classic LoRA failure: a LoRA that does nothing usually means you forgot its trigger word, not that the node is broken.

Install

Install once for the whole pack, since all four nodes ship together:

cd ComfyUI/custom_nodes
git clone https://github.com/drago87/ComfyUI_Dragos_Nodes

Then restart ComfyUI. Or use ComfyUI Manager and search "ComfyUI_Dragos_Nodes". No pip dependencies, no model downloads - it only uses ComfyUI core, so install is instant. It's a one-commit, personal utility pack from late 2023 and hasn't been touched since, so temper expectations: the code has rough edges and the README has typos, but for the narrow job of "load a LoRA and tell me its name and strength" it works.