⭐ Star Ollama Prompt Helper

You know the loop: stuck on a blank prompt box, so you open ChatGPT, ask it to describe a cyberpunk lantern festival, paste the result back. Star Ollama Prompt Helper removes the middleman - it runs the same trick against a local Ollama model, inside your workflow, and hands the result straight to your prompt encoder. No cloud, no API key, no copy-paste, and (privacy aside) it works offline.

The README frames Ollama as an optional add-on, and that's the right framing: the node needs the Ollama server running and a model pulled, but once that's true it's genuinely useful for prompt iteration. It even has vision support - plug an IMAGE into the optional input and reference it as "this image" in your prompt, which turns it into a crude captioner/refiner for whatever you just generated.

How it works

The node talks to your local Ollama server over HTTP (default http://127.0.0.1:11434) using the ollama Python package, which is a real dependency in the pack's requirements.txt. The model dropdown starts empty until you hit refresh, at which point it lists the models your server has pulled. system_prompt_preset pulls from the pack's shared text storage (json/startext.json) - 19 presets covering things like the Krea2 image prompt refiner, Flux Klein edit, and LTX prompt refiners - or you pick Custom and write your own system prompt.

allow_thinking is the interesting control: with reasoning models, set it true to get the model's chain-of-thought back on the think_output socket, or false to tell the model to skip reasoning and answer faster.

Inputs that matter

• local_address - your Ollama server. If you run Ollama in Docker or on another machine, change this.
• model - refresh to populate from your server.
• system_prompt_preset - the preset list, or Custom + your own system_prompt.
• prompt - what you're asking the model to do with your image-generation prompt.
• temperature - creativity dial (default 0.8).
• keep_alive - how long the model stays hot in memory: -1 forever, 0 unload immediately (slow on repeat runs), defaults to 5m.
• image - optional vision input; refer to it as "this image."

Outputs: result (the model's answer - wire into your CLIP text encode) and think_output (the reasoning trace, when enabled).

Install

Install the pack as usual:

cd ComfyUI/custom_nodes
git clone https://github.com/Starnodes2024/ComfyUI_StarNodes
cd ComfyUI_StarNodes
pip install -r requirements.txt

Then make sure the ollama package landed (it's in requirements) and - the step people forget - start the Ollama server and pull a model:

ollama serve
ollama pull llama3.2:latest

Restart ComfyUI, search star. If the node shows no models, refresh the dropdown with the server running.

Where people get burned

"Node can't connect" is almost always "Ollama isn't running" - the README's troubleshooting table says it straight: start the app/server first. Second, vision models and text models aren't the same thing: plugging an image in while using a non-vision model just wastes the input. And the presets live in the pack's shared text storage, so if you edited startext.json and your custom preset vanished, it's a storage-path issue, not a node bug.