OobaPrompt (Mikey)

Same idea as Mikey Nodes' LM Studio Prompt node, different backend: this one talks to oobabooga's text-generation-webui, the other big name in local LLM UIs, the one that's been around since the early Llama days and grew its own sprawling extension ecosystem the way Automatic1111 did for image generation. If you're already running text-generation-webui for chatting with local models, this node lets you point that same setup at prompt generation.

What it does

You type a rough idea into input_prompt, pick a mode, and the node sends it off to your running Oobabooga instance and returns generated prompt text. The mode list here is close to, but not identical to, its LM Studio sibling: prompt, style, descriptor, character, negative, custom - six presets, each almost certainly triggering a different built-in system prompt tuned for that kind of output. The negative mode stands out as the one option this node has that LM Studio Prompt doesn't (which has pixelwave instead) - worth trying if you want the LLM to specifically generate negative-prompt text rather than positive description.

custom_history works the same way it does on the LM Studio node: point it at a JSON file containing your own conversation history or system prompt, and pick custom mode to have the node use that instead of any built-in preset. seed almost certainly drives the LLM's own sampling, so the same input at the same seed should give you a reproducible result.

One notable difference from its sibling

The schema for this node has no server_address or server_port fields the way LM Studio Prompt does. I can't verify from the schema alone whether that means the connection details are hardcoded to Oobabooga's usual defaults, configured somewhere else in the pack, or handled differently altogether - if you're setting this up and it's not reaching your text-generation-webui instance, that's the first thing worth digging into rather than assuming the node is broken.

Output

One text STRING - feed it straight into a CLIP Text Encode node, or chain it through Mikey Nodes' Wildcard Processor or LoraSyntaxProcessor first if you want the LLM's output to still resolve wildcard tags or inline <lora:> syntax.

Installing

Comes with the full Mikey Nodes pack - ComfyUI Manager, search "Mikey Nodes," install, restart, or cd ComfyUI/custom_nodes && git clone https://github.com/bash-j/mikey_nodes and restart manually. The node itself needs nothing extra from the pack side. What you need separately is oobabooga's text-generation-webui installed, a model loaded, and its API extension enabled - text-generation-webui ships an OpenAI-compatible API as an optional extension, and it typically needs to be explicitly turned on before any external tool (this node included) can reach it.

Where it breaks

• No response, or a connection error. The most common cause by far: text-generation-webui's API extension isn't enabled, or the server simply isn't running. Check the Oobabooga window/logs before troubleshooting anything on the ComfyUI side.
• Output that ignores the chosen mode. If the model loaded in text-generation-webui isn't a proper instruction/chat-tuned model, it may not follow the preset's system prompt reliably - a raw base model will produce much weaker results here than something like an instruct-tuned Llama or Mistral variant.
• Reasoning text leaking into the output. If you're running a reasoning-style model through this node, its raw response can include a <think>...</think> block ahead of the real answer. Chain the pack's own Remove Text Between node right after this one (its defaults are already set to strip exactly that tag pair) if you see that happening.