Random LoRA Model Loader (Compact)

LoRA exploration has a dirty secret: half the work is remembering which trigger words go with which LoRA. Random LoRA Model Loader (Compact) is AUN's answer - a single node with up to 10 LoRA slots, each carrying its own trigger words, that rolls a slot (Select, Increment, Random, or Range), applies that LoRA to your model, and hands you the trigger text ready to drop into your prompt. It's the random-selection take on the idea rgthree's Power Lora Loader popularized, with the trigger management baked in.

How it works

You load up to 10 lora_N slots (dropdowns populated from your loras folder), each with a trigger_N string. A mode picks how the active slot is chosen - Select (fixed select), Increment/Random (within minimum/maximum), or Range (1,3,5-6 style list). On run it applies that LoRA to the incoming model using strength_model, outputs the patched MODEL and CLIP (the CLIP gets strength_clip), and returns everything you need for prompt assembly.

The outputs are where this shines:

• MODEL / CLIP - the LoRA-applied pair (CLIP only meaningful when you connect the optional clip input; without it, CLIP passes through).
• selected LoRAs (STRING) - the name of the active LoRA.
• trigger words (STRING) - the active slot's trigger text.
• trigger + prompt (STRING) - triggers plus the optional base_prompt text, ready to append to your prompt or chain into another prompt builder.
• index (INT) and label (STRING) - which slot won and its label, for downstream control or filenames.

Because trigger words usually need to be in the prompt for the LoRA to work, the base_prompt optional input is the smart touch: wire your normal prompt in, and trigger + prompt gives you the combined string - one wire into CLIP Text Encode instead of manual pasting.

The inputs that matter

• mode, select, minimum, maximum, range - the selection logic (same family as the pack's index nodes).
• lora_N / trigger_N - the slots themselves. This is where you do the setup work.
• strength_model / strength_clip - LoRA strengths. strength_clip only does anything when a CLIP is connected.
• apply_lora - master switch; off = returns the input model unchanged. Great for A/B "is the LoRA even helping" checks.

Compact mode, and the workflow it enables

Double-click the body (or right-click → Compact mode) to hide the config widgets; in compact mode the footer shows the active slot's trigger words, and you can hide/show clip strength and the footer from the right-click menu. Pair it with a PromptCycler or a Random/Select INT for the index, and you get rotating LoRA+prompt combos in one clean column.

Install

Part of AUN ComfyUI Nodes (loz2754):

cd ComfyUI/custom_nodes
git clone https://github.com/loz2754/AUN-ComfyUI-Nodes

or ComfyUI Manager → "AUN ComfyUI Nodes" → install → restart. Pack deps (piexif, opencv-python-headless, imageio-ffmpeg, requests) via Manager; no model downloads - it just reads whatever LoRAs are already in your ComfyUI/models/loras folder.

Gotchas

Trigger words only work if you actually use the trigger + prompt output - if you send your own static prompt to CLIP Text Encode, the triggers never land. And when the pack rolls a None slot (empty lora_N), it applies nothing; if a slot's LoRA seems to vanish, check whether that slot is filled. For stacking multiple LoRAs per prompt, the pack's Random Multi-LoRA Model Loader is the 20-prompt×3-slot version.