🦙 Ollama Configuration ⚙

Bjornulf's Ollama Talk and Ollama Vision nodes work fine untouched - they default to llama3.2:3b on http://0.0.0.0:11434 for chat, and moondream for vision. This node exists for the moment you outgrow those defaults: a bigger local model, a different Ollama host, or a vision model that also happens to be decent at plain text. Plug it into either node's optional config slot and it overrides both the URL and the model in one place.

How it works

It's a small bundling node with no logic of its own - you fill in an Ollama server URL and a model name, and it packages both into a single OLLAMA_CONFIG object that Ollama Talk and Ollama Vision both accept as an optional input. Without it connected, those nodes fall back to their hardcoded defaults; with it connected, whatever you type here wins.

Worth being explicit about since it trips people up: Ollama itself has to already be installed and running for any of this to work - these nodes are a thin client that talks HTTP to a real Ollama server, they don't bundle or launch one. If you haven't set that up, grab it from ollama.com and pull whatever model you plan to point model_name at (ollama pull llama3.2:3b, or whatever vision or chat model you want) before wiring any of this up.

The inputs and outputs that matter

• ollama_url (default http://0.0.0.0:11434) - where your Ollama server is listening. Only change this if you're running Ollama on another machine or a non-default port.
• model_name (default unset) - the exact model tag as Ollama knows it, e.g. llama3.2:3b, moondream, llava. It needs to already be pulled locally, or the request just errors.
• Output: OLLAMA_CONFIG - feed this into the optional config input on Ollama Talk or Ollama Vision.

How to install it

Via ComfyUI Manager, search Bjornulf_custom_nodes and install. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/justUmen/Bjornulf_custom_nodes

Then from inside that folder, pip install -r requirements.txt (a venv is worth the extra step here). One dependency actually matters for this specific node family: pip install ollama pulls in the Python client the pack talks to Ollama's HTTP API with. The README is explicit that you don't have to install the Ollama application itself unless you want to actually use these nodes - but the ollama Python package has to be installed regardless, because the custom node file imports it unconditionally. Skip it and none of the ~170 nodes in this pack will load, not just the Ollama ones.

Common issues

The single most common failure mode with any of these Ollama nodes is a plain connection refused - because either Ollama isn't running, or ollama_url doesn't actually point at it. The default http://0.0.0.0:11434 assumes Ollama is running on the same machine as ComfyUI; if you're running ComfyUI in a container, on a remote box, or through a cloud GPU rental, 0.0.0.0 almost never resolves to the right place and you'll need to point this node at wherever Ollama is actually reachable.

The second most common one is a model-not-found error - model_name has to match a model you've already pulled with ollama pull, exactly, tag included. Ollama doesn't download models on demand for these nodes; if it's not already in your local ollama list, the request fails.