🌐 Remote Text Model Selector

If you run Ollama locally, you already have a small LLM that ComfyUI would love to borrow. This node is the bridge: it looks at your running Ollama (or any OpenAI-compatible service), lists the models, and hands the pack a ready-to-use remote model config. Note the word selector - nothing generates here. The actual text comes out of the pack's TextGeneration node, which this feeds.

Ollama-behind-ComfyUI is a well-trodden pattern, by the way. People have been wiring a quick LLM pass to enrich a doodle prompt before it hits the KSampler since 2024, and it's one of the most common ways ComfyUI users pull a text model into the graph. This node just turns that from a script into a dropdown. And the name's no lie: it's not calling a paid API and needs no key - it's talking to your own service on port 11434.

How it works

When the node loads, it polls http://127.0.0.1:11434/api/tags (it tries port 11435 as a fallback) and builds the model dropdown from whatever Ollama reports. That list is frozen at load time, so new ollama pulls won't show until you restart ComfyUI - the classic gotcha, and the reason "No models found" is the default enum entry when nothing's running at startup.

api_type switches the backend between three options:

• Ollama - the recommended default, and the one that auto-populates the dropdown
• Nexa SDK - Nexa's local service, same OpenAI-compatible idea
• OpenAI Compatible - any server speaking the OpenAI protocol (LM Studio, vLLM, whatever)

The output is a TEXT_MODEL config - base_url, selected model name, and the optional system_prompt bundled together. Wire it into the pack's TextGeneration node, which shares the same TEXT_MODEL type, and you're generating.

The inputs that matter

• base_url - http://127.0.0.1:11434 by default; change it if your Ollama is elsewhere.
• api_type - leave on Ollama unless you're targeting another backend.
• model - the dropdown, auto-filled from the service. This is the one you actually care about.
• system_prompt - optional; the natural home for "expand this into a full prompt, then answer with one paragraph."

Common issues

No models found in the dropdown is almost always one of three things: Ollama isn't running (ollama serve), it has nothing pulled yet (ollama pull llama3.2), or you added a model after ComfyUI started and need a restart. Check the service first, restart second. If you're hitting the wrong port entirely, fix base_url.

Install

ComfyUI Manager (search "ComfyUI-GGUF-FX") or:

cd ComfyUI/custom_nodes
git clone https://github.com/weekii/ComfyUI-GGUF-FX.git
cd ComfyUI-GGUF-FX
pip install -r requirements.txt

Restart, make sure Ollama is up, and the dropdown should fill itself. This is the pack's lightweight remote path - no model downloads through ComfyUI, no heavy transformers stack required, just a running service and a node that knows how to find it.