Scorpiov Wildcard Processor 🎲

Wildcards are how you turn one prompt into a thousand. The idea is old - it's A1111-era dynamic-prompt syntax that most people arriving after 2024 never learned - and it's simple: write a __hairstyles__ woman, {studio|street|golden hour} and each run rolls a different hairstyle from a file and a different setting from the inline group. Scorpiov Wildcard Processor is the all-in-one version: it resolves wildcards, resolves inline {a|b|c} groups, loads any <lora:...> tags it finds, and outputs the finished CONDITIONING - plus the model, clip, and resolved text. One node replacing the usual "wildcard node → CLIPTextEncode → LoraLoader" tangle.

How it works

The inputs are the giveaway that this is the heavyweight variant: it takes model and clip directly, because it can load LoRAs itself and it encodes the conditioning itself. The full set:

• text - your prompt template, with __file__ references and {a|b|c} groups (nested groups supported).
• mode - random (seeded) or serial. Serial remembers position and loops through the file, which is what makes the Batch Prompt Runner tick.
• seed - controls random-mode picks.
• serial_start_line - which line serial mode starts from (1 = first), applying to all wildcards in the prompt.

Wildcard .txt files live in the pack's wildcards/ folder, referenced by filename only (__hairstyles__, no path) - any subfolder depth works. Comments are supported both in .txt files and in the node's text box: # to end of line, and C-style /* ... */ blocks that can span lines. They're stripped before any wildcard resolution, so they never reach the resolved prompt, Save Image, or embedded metadata. There's a live 🎨 Highlighted Preview under the text field, color-coding groups, weights, numbers, and comments as you type - read-only, just a visual aid, but a genuinely nice one for debugging a gnarly template.

Outputs: conditioning (ready for a KSampler), the LoRA-applied model and clip (so downstream loaders can chain), and processed_text (the resolved string, handy for display or for wiring into Save Image's prompt inputs).

Where it sits

This is the "just do the whole thing" node. The text-only sibling - Wildcard Prompter - deliberately has no model/clip inputs and only outputs the string, for when you want to resolve text first and decide later. The Processor is for when you want the pipeline collapsed: template in, conditioning out, LoRAs along for the ride.

Two implementation details worth knowing. IS_CHANGED returns NaN, which tells ComfyUI this node must re-run every time - that's what makes random mode actually roll fresh picks. And the resolved text is printed to the server console on each run, which is either a debugging gift or log noise depending on how much you batch.

Gotchas

• The wildcards folder lives inside the pack folder, not in your main config. Drop files in ComfyUI/custom_nodes/scorpiov-nodes/wildcards/. There's a "🔄 Refresh Wildcards" button on the node that rescans the folder and resets serial state - use it after adding files, because the index is built once and cached.
• A missing wildcard file resolves to its literal __name__ text with a log warning, so a typo shows up as a prompt that contains __hairsyles__ as plain words. That's the classic silent failure; the highlighted preview helps you catch it before you queue.
• LoRA tags are loaded at full pack strength - wire the node's model/clip outputs onward, not the originals, or the LoRAs silently drop.

Installing

Part of scorpiov-nodes. Install via ComfyUI Manager (search "scorpiov-nodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/vikramudyawar/scorpiov-nodes

Restart ComfyUI after. No extra Python dependencies, no models to download - just the wildcard .txt files you bring. If you've been hand-rolling {a|b|c} prompts in your head, this is the node that makes batch variety one template away.