Load LoRA Tag [LP]

If you came to ComfyUI from Automatic1111 or Forge, you probably miss typing <lora:mylora:0.8> directly into your prompt and having it just work, instead of wiring up a separate LoRA loader node and threading MODEL/CLIP through it by hand. This node gets you that back.

What it does

This parses Automatic1111/Forge-style LoRA tags straight out of a text field - the <lora:name:weight> syntax that ecosystem standardized on - and applies each one to your model and CLIP in sequence, no separate loader node per LoRA needed. Instead of daisy-chaining three LoraLoader nodes for three LoRAs, you write all three tags into one text box and this node handles the rest. The README credits the core functionality to comfyui_lora_tag_loader by badjeff, with LevelPixel's version adding a text output for load errors as a widget on the node - so if a tag references a LoRA file that doesn't exist or fails to load, you get that surfaced on the node itself rather than a silent skip or a buried console error.

This is a genuinely common want in the ComfyUI community - people managing LoRA stacks by editing tag text directly, or driving randomized/wildcard LoRA selection by generating the tag string dynamically upstream and feeding it into a loader exactly like this one, rather than rewiring nodes every time the LoRA combination changes.

Inputs and outputs

• model (MODEL) and clip (CLIP) - your base model and CLIP, same as any standard LoRA loader.
• text (STRING, multiline) - where the <lora:name:weight> tags live, mixed in with the rest of your prompt text if you want.

Outputs:

• MODEL, CLIP - the same model and CLIP with every LoRA tag found in text applied, in sequence.
• STRING - the pass-through text output.

How to write the tags

The syntax follows what Automatic1111 and Forge use: <lora:filename:weight>. Multiple tags in the same text box get applied in order. Watch your weights - the old rule of thumb (0.5–0.8) is SDXL-era advice and doesn't generalize; newer architectures run differently, with some Z-Image character LoRAs pinned at 1.0 or higher because dropping the weight loses likeness. Check what the specific LoRA's page recommends rather than defaulting to a number you remember from a different model family.

Installing it

ComfyUI Manager: search "ComfyUI-LevelPixel" (listed as "Level Pixel"), install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/LevelPixel/ComfyUI-LevelPixel.git, restart ComfyUI. No models download automatically - you still need the LoRA files themselves in your usual LoRA folder, same as any other loader; this node just changes how you reference them.

Common issues

The most common failure mode with any tag-syntax LoRA loader is a typo or mismatch between the tag's filename and what's actually on disk - the filename inside <lora:...> needs to match your LoRA file exactly (minus extension), and this pack's version specifically surfaces that mismatch as an error widget on the node rather than leaving you to dig through logs, which is worth checking first if a LoRA you referenced doesn't seem to be applying.

Architecture mismatches are the other classic trap, and it's not specific to this node - any LoRA trained for one base model family (SDXL, Flux, Illustrious) will not do anything useful, or will actively degrade output, if applied to a checkpoint from a different family. If your LoRA tag loads without error but the effect looks wrong or absent, check the LoRA's listed base model against what you're actually generating on before assuming the node is at fault.

And if you're stacking several LoRAs through this one text box, remember interactions between them are genuinely unpredictable - add them one at a time with a fixed seed if something looks off, so you can tell which tag is actually causing the change.