Boyo LoRA Config Processor

The pack splits its LoRA machinery into two halves on purpose. One half loads LoRAs and patches models; the other half just does the bookkeeping - reading configs, resolving prompts, and producing clean path data. Boyo LoRA Config Processor is the bookkeeping half. It reads up to three LoRA configs, picks a prompt from each according to a prompt mode, applies a per-slot strategy, and emits the resolved config data plus prompt strings. You feed its config_data outputs into the pack's LoRA Path Forwarder, which turns them into stable LoRA paths for standard ComfyUI loaders.

Understanding the two knobs is understanding the whole tool:

• prompt_mode - how to choose a prompt from each config's prompt list: First Only, Cycle Through (advance through the list), or Random (seeded, so reproducible).
• prompt_strategy_1/2/3 - what to do with a slot's prompt: Mute drops it (great for utility LoRAs that shouldn't pollute the prompt), Concatenate keeps it, and Merge is currently implemented the same as Concatenate - the source has a TODO for smarter merging. So don't expect Merge to do something magical yet.

The inputs

• prompt_mode - the selection strategy, default First Only.
• prepend_text / append_text - text wrapped around the final combined prompt. Handy for a fixed quality tag or style suffix.
• lora_config_1/2/3 - dropdowns of config files; None for empty slots.
• prompt_strategy_1/2/3 - per-slot Mute/Concatenate/Merge.
• seed - for Random mode, default 0.
• refresh_configs - BOOLEAN; tick it to rescan the configs folder without a restart.

Outputs: config_data_1/2/3 (DICTs carrying each config's resolved data - these plug into Boyo LoRA Path Forwarder), prompt_1/2/3 (each slot's resolved prompt), and combined_prompts (prepend + joined prompts + append).

Install

Part of Boyonodes. ComfyUI Manager → search "Boyonodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/DragonDiffusionbyBoyo/Boyonodes

Restart. No extra dependencies.

The one mental shift

This node never touches a model. If you wire its STRING outputs into a CLIP text encode and expect LoRAs to load, nothing happens - that's the Paired Loader or Path Forwarder's job. The Processor is a data node; its outputs are for other nodes to consume. Get the split right and the toolset clicks: build configs with JSON Builder, inspect them, process prompts here, and forward the paths downstream.