LoRAs by Prompt Index

The classic multi-prompt LoRA problem: you're cycling through prompts, and each one wants a different LoRA stack. Prompt 1 needs the photoreal skin LoRA, prompt 2 needs the stylized one, and they definitely don't want each other's trigger words. Doing that by hand means re-patching the model between runs. AUNLoRAsByPromptIndex is the automated version: a prompt_index input decides which of up to 20 prompts' LoRA sets gets applied, so a prompt cycler upstream swaps the entire stack for you on every run.

Mechanically it's exactly what it says. You get up to 20 prompts, each with 3 LoRA slots. Each slot has pN_lora{S} (the file), pN_strength_model{S}, pN_strength_clip{S} (if you connect a CLIP), pN_trigger{S} (the trigger words), and pN_enabled{S}. At execution it reads prompt_index, grabs that prompt's slots, and applies the active LoRAs sequentially to the same model+clip. There's no built-in randomizer here - the node is purely index-driven, so Select/Range/Random behavior comes from whichever upstream node feeds prompt_index (an AUN Text Index Switch's index output, an AUNMultiPromptCycler, whatever you like). That's the architecture: one node owns the index, this one owns the LoRAs.

Outputs: MODEL (patched), CLIP (patched if connected, else passthrough), selected LoRAs (a STRING of <lora:name:strength_model:strength_clip> tags), index, trigger words, and trigger + prompt (trigger text combined with the optional base_prompt input). Those string outputs are gold if you want the actual active LoRA tags or triggers visible somewhere, or concatenated into chained stacks via the selected_LoRAs pass-through input.

The UI does the heavy lifting for sanity: empty slots are hidden - a prompt with no LoRAs shows no slot rows at all, so you only ever see what's configured (purely cosmetic; execution skips None/disabled slots anyway). Double-click for compact mode, where you can drag one LoRA label onto another to swap their values, enable/disable per slot, and see a footer of combined trigger words. A full-screen Setup dialog handles editing any prompt regardless of what's visible.

This is the recommended successor to the older AUN Random Multi-LoRA Model Loader - same behavior, minus the empty-slot noise.

Install: ComfyUI Manager (search "AUN"), or cd ComfyUI/custom_nodes && git clone https://github.com/loz2754/AUN-ComfyUI-Nodes, restart. If a manual install hits missing modules, pip install -r custom_nodes/AUN-ComfyUI-Nodes/requirements.txt.

Troubleshooting:

• Nothing changes when you flip prompt_index → check apply_lora isn't off (off returns the model unchanged, by design).
• Clip strength inputs do nothing → there's no CLIP connected; they only matter with one.
• Slots go up to strength ±20, and negative strengths are legal (they weaken/attenuate a LoRA) - don't panic if you see negatives.
• Trigger words do nothing until you actually put them in your prompt; they're strings, not magic.