Wildcard OobaPrompt (Mikey)

The README groups this one with OobaPrompt and LMStudioPrompt under a single line: "prompt processing nodes that integrate with external LLM services." The "Wildcard" prefix tells you what makes this one different from plain OobaPrompt - it resolves this pack's __wildcard__ syntax in your prompt first, then sends the result to a local LLM running through Oobabooga's text-generation-webui for further processing, and hands you back whatever the model produces.

Why chain wildcards into an LLM at all

This is genuinely a real, community-established pattern, not something bash-j invented in isolation - other ComfyUI node packs (Plush for ComfyUI is a well-known example) offer the same idea of routing a prompt through a local LLM front end like Oobabooga or LM Studio before it hits your image model, specifically to expand a short idea into a fuller, more descriptive prompt, or to rewrite tag-soup into natural language for models that want that. Stacking wildcards underneath that is a natural extension: your wildcard file gives you random variation in the raw ingredients (a character trait, a setting, a mood), and the LLM turns whatever combination got picked into a coherent, well-formed prompt instead of you writing the glue text by hand every time.

What you set

• input_prompt - your starting text, wildcards and all. Anything using the pack's __word__ syntax (including the 2$$, |word filter, and !/+/-/* selection-locking variants documented in the pack's README) gets resolved before the LLM sees it.
• seed - controls which wildcard lines get picked when there's randomness involved, so a given seed reproduces the same starting text run to run.

Output is a single text string - the LLM's response, ready to feed straight into your CLIP Text Encode node.

Installing it - and what it actually needs to run

The node itself installs the normal way, through ComfyUI Manager (search "Mikey Nodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/bash-j/mikey_nodes

Restart ComfyUI afterward. But unlike most of this pack, this specific node has a real external dependency the README doesn't spell out in detail: a running instance of Oobabooga's text-generation-webui, launched with its API flag enabled, and a model loaded into it. That's a separate piece of software you install and run yourself - it's not bundled with Mikey Nodes, and ComfyUI won't set it up for you. text-generation-webui's API listens on port 5000 by default; if this node isn't finding your instance, that's the first thing to check.

Where this goes wrong in practice

The most common failure is the boring one: Oobabooga isn't running, isn't running with its API enabled, or doesn't have a model loaded - any of which will leave this node with nothing to talk to. Confirm you can hit the API directly (or at least confirm the webui's own interface is up) before troubleshooting inside ComfyUI.

Since the node's exact connection settings (host, port, any auth) aren't documented in the pack's README, and this article's briefs don't expose them either, budget time to check the node's widgets directly in ComfyUI once it's loaded - this is one of the less-documented corners of the pack, grouped with two other LLM-integration nodes that get one line of description each rather than a dedicated section.

If your wildcards aren't resolving before the LLM sees them, make sure your wildcard files are sitting in ComfyUI/wildcards/ with filenames matching exactly what's between the double underscores - that's the same rule as everywhere else this pack's wildcard engine shows up, and a mismatch here fails silently rather than erroring.