Ollama LLM Connector 🦉| OwlV

Why you'd reach for it

Ollama is the easiest way to run local LLMs that exists: one binary, ollama pull llama3, done. So an "Ollama LLM Connector" in a translation pack looks like the perfect bridge - free, local, private translation inside ComfyUI, no API key, no billing. That's the draw, and it's a good one. The catch is what actually shipped.

How it actually works

The connector takes a host (default http://localhost:11434) and a model (default llama3:8b), builds an LLM client, and… stops. The client's translate() returns a string shaped like [Ollama http://localhost:11434/llama3:8b] -> 中文: your text without making a single HTTP request to Ollama. The source has the TODO comment where the API call should go, and a placeholder body instead. Hook it into the LLM Translator node and you get a string that looks like a confirmation but isn't a translation.

There's also an api_key field, password-masked. It's harmless but pointless - stock Ollama doesn't use API keys. It's leftover from the pack's service-connector template.

The inputs

• host - your Ollama server; default http://localhost:11434.
• model - any model you've pulled; default llama3:8b.
• api_key - optional, unused in practice.
• system_prompt - optional; defaults to the pack's "You are a translation engine. Only output the translated text." prompt.

Output: llm, which plugs into LLM Translator's llm input.

Installing it

Ships in the ComfyUI-MultiTranslator pack by OwlvChirotha ("OwlV"). ComfyUI Manager → search "ComfyUI-MultiTranslator" → Install → restart. (The README's "ComfyUI-Translator" search string is stale.) Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/OwlvChirotha/ComfyUI-MultiTranslator
cd ComfyUI-MultiTranslator
pip install -r requirements.txt

Restart, then look under Text Processing/LLM. Dependencies are just googletrans and requests - no ollama Python client, another tell that the integration isn't wired up.

The honest verdict

If you want Ollama-based translation in ComfyUI today, this node can't deliver it - yet. The groundwork (host/model plumbing, the llm output type, the system prompt) is all there, and a real implementation would slot straight into the existing interface. Until then, either use a different pack's Ollama node - several do real inference - or call Ollama's API directly. Watch this pack for an update; it's young, and this is the most likely piece to get finished next.