⚙️ FMJ Llm Config

The most useful node in the FMJ-LLM pack is also the most boring one, and that's the point. ⚙️ FMJ Llm Config doesn't generate anything. It doesn't call a model, doesn't need an API key, doesn't touch your images. It's a config passthrough: you pick your Ollama URL and your model once, and it hands those two strings to every other node in the pack. If you're running several FMJ nodes in one workflow, it saves you retyping qwen3:2b and http://localhost:11434 in every single one.

Here's the part that makes it worth a look: the model drop-down isn't a hardcoded list. When the node loads, it asks your Ollama server for GET /api/tags and builds the selected_model menu from the models you've actually pulled. So instead of guessing what's installed, you pick from a menu of what's there. If Ollama isn't reachable at that moment, it falls back to a placeholder list - llama3, qwen3, moondream, llava, phi3 - so the node still renders and you can fix the connection later.

The inputs and outputs that matter

There are only two inputs and they're both obvious:

• ollama_url - default http://localhost:11434. Point it at your Ollama server, or a remote one if you're running Ollama on another box.
• selected_model - the dropdown, populated from /api/tags at node-creation time.

It returns two STRING outputs: model_name and ollama_url. Wire those into the model_name and ollama_url inputs of the Prompt Generator or the Vision node and you've got one central place to change your model. Nothing else happens - it's literally a string in, two strings out.

One honest caveat: the model list is fetched once, when the node is created, using the default URL. If you change ollama_url on an existing node or pull new models later, the dropdown won't refresh itself - delete the node and add it again, or restart ComfyUI.

Installing it

It's part of the bulldog68/ComfyUI_FMJ_LLM pack, so you install all four nodes at once:

• ComfyUI Manager (easiest): search "ComfyUI_FMJ_LLM" or "FMJ-LLM" and install.
• Manually: cd ComfyUI/custom_nodes && git clone https://github.com/bulldog68/ComfyUI_FMJ_LLM, then restart ComfyUI.

The only Python dependency is the ollama package, which Manager pulls for you. You do need Ollama itself installed first (that's the separate ollama.com runtime) - without it, the node just shows you the fallback model list.

Troubleshooting

• Dropdown stuck on the five fallback names - Ollama wasn't running when the node loaded. Start Ollama, then recreate the node. You can also just type a model name; it's a plain string under the hood.
• Nothing to fix? Then skip this node entirely. The other FMJ nodes all have plain text model_name fields, so this is a convenience, not a requirement. It only earns its keep when you've got several LLM nodes in one graph.

For a "config" node there's not much more to say, and that's a compliment. If you're using the rest of the FMJ pack, drop this in, set it once, and forget it.