LoRA Loader 🆎

The default ComfyUI experience with LoRAs is a row of loader nodes, each with a strength slider, wired in series - and if you use three or four of them, the graph is a mess and switching combinations means rewiring. LoRA Loader 🆎 is the AusBoss pack's collapse of that whole pattern: one compact node with a row editor for the entire stack.

How it works

Each row is a LoRA with three things: an on/off pill, a searchable picker, and model/CLIP strengths. The picker is worth calling out - type to filter, arrow keys + Enter to pick, with a browse view that groups by folder and shows the preview image on hover. The strengths are the fun part: drag left/right to scrub (Shift for fine steps) or click to type. It's the kind of interaction that makes you stop using every other LoRA loader.

Rows apply in order: model and clip outputs carry the whole stack applied in row order. clip is optional - leave it unwired and CLIP strengths are simply ignored. A header pill toggles the entire stack at once, which is the "compare with and without all LoRAs" button you didn't know you needed.

The outputs

model - the diffusion model with every enabled LoRA applied.
clip - the CLIP with every enabled LoRA applied (patched only if a CLIP input was connected).
trigger_words - comma-joined trigger words from all enabled rows, deduplicated. This is a genuinely useful output that most stack nodes skip: the row's info card reads each LoRA's metadata (preview image, base model, trigger words from the file), so you can click words to toggle them into the trigger_words string - or save your own. One cable into a prompt node and the trigger words are always right for the stack you have enabled. The trigger_separator (default ", ") is hidden on the canvas, set from the gear menu.

The bar's ▤ button saves and applies named templates of the whole stack, so a character's three-LoRA combo with tuned strengths is one click to reload. And a nice safety net: a LoRA that patches nothing on the connected model - the usual sign it was built for a different base model - logs a console warning naming it. The KB's rule applies: architecture incompatibility is absolute; a Flux LoRA on an SDXL model does nothing, and this node tells you when that's happening.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/ausboss/ComfyUI-AusBoss.git

Restart ComfyUI and search for AusBoss (Manager: ComfyUI-AusBoss). No extra Python dependencies; minimum ComfyUI 0.27.1.

Troubleshooting

A "patches nothing" warning is your strongest signal - either the LoRA is for a different base model or the model input isn't what you think. If trigger words come out empty, the LoRA file has no trigger-word metadata; that's when you save your own words in the info card. And if a stack mysteriously applies with half the rows ignored, check the enable pills - the header toggle turns everything off at once, and it's easy to hit it accidentally.

Name	Type	Default	Description
model	MODEL		The diffusion model the LoRA stack patches.
loras	STRING	[]	Serialized LoRA stack managed by the node's row editor; edit the rows, not this value.
clipopt	CLIP		Optional CLIP to patch; without it, CLIP strengths are ignored.
trigger_separatoropt	STRING	,	Joins the enabled rows' trigger words into the triggers output. Set from the node's gear menu; hidden on the canvas.
on_missingopt	COMBO	error	What happens when an enabled row's file cannot be found. error (the default) fails validation before sampling and names the file, so an unattended or scripted run never renders without a LoRA it asked for; skip warns once in the console and applies the rest of the stack. Set from the node's gear menu (Stop on missing LoRA); hidden on the canvas.

Name	Type	Description
model	MODEL	The model with every enabled LoRA applied in row order.
clip	CLIP	The CLIP with every enabled LoRA applied in row order.
triggers	STRING	Comma-joined trigger words from enabled rows, deduplicated.

How it works

The outputs

Install

Troubleshooting

Inputs (5)

Outputs (3)