LoraX

Most ComfyUI workflows end up with two or three stacked Lora Loaders, each holding one LoRA and its strengths, and the order they apply in is a guessing game buried in the wire layout. LoraX is WorkflowX's answer: one node with a visible, reorderable stack of LoRA rows - add several, drag them into order, give each its own model and CLIP strength, and it applies them in exactly that sequence. It's the same job as rgthree's Power Lora Loader, done the WorkflowX way, and honestly both are a big step up from chaining single-LoRA loaders.

The stack isn't just for show. The node applies enabled rows in displayed order (order matters for LoRAs that fight over the same concepts), skips empty rows, and - this is the part that saves real pain - reports unavailable filenames instead of silently substituting. If you renamed a file in your models/loras folder, LoraX tells you the row broke rather than quietly loading something you didn't pick.

Inputs and outputs

• model (MODEL) - your base diffusion model. Required.
• clip (CLIP) - optional. Feed it if you want the LoRA stack to patch the text encoder too.

The LoRA rows are UI-managed: add, remove, reorder, search, select, enable/disable, refresh. Each row is a searchable picker over your LoRA folder, with separate model and CLIP strength fields. The picker leans on LoRA Manager's metadata and previews when you have that pack installed, but works fine from plain folders alone.

Outputs:

• MODEL - the model with enabled LoRAs applied in order.
• CLIP - the patched text encoder, when one was supplied.
• trigger_words (STRING) - trigger words gathered from the enabled rows. Handy if you want to dump them into a prompt or a note node automatically.
• loaded_loras (STRING) - a human-readable audit of what actually got applied. Wire this to a text display when you're debugging why a stack behaves oddly.

Install

The pack is one install:

cd ComfyUI/custom_nodes
git clone https://github.com/haroonaslam/WorkflowX-Configurator

or search WorkflowX Configurator in ComfyUI Manager, restart ComfyUI, refresh the browser. No pip dependencies, no model downloads - the only "dependency" worth knowing is that LoRA Manager is optional and only enriches the picker previews.

Where people get burned

• Order is not cosmetic. LoRAs apply in displayed row order, and reordering mid-workflow changes results. If two LoRAs both touch a concept, the later one tends to win the tug-of-war.
• Strengths are per-row and split. A row has separate model and CLIP strength. Forgetting to set CLIP strength when the LoRA needs it is the classic "why is this LoRA doing nothing" - the model is patched, the text encoder isn't.
• Unavailable files fail loudly, on purpose. That's a feature - it caught the renamed-file case. But it means you can't run a workflow while a LoRA is missing; fix the row or disable it.
• It's not the same as trigger words working. The node gathers trigger words for you, but pasting them in is still your job - and for modern LLM-encoded bases, natural-language trigger phrasing beats tag-style.

The one LoRA node to reach for if you routinely stack more than one LoRA and want to see exactly what's applying, in what order.