Ollama Image List Connectivity

"Ollama Image List Connectivity" sounds like the least exciting node in a pack that already isn't flashy, but it solves a genuinely annoying problem: the Generate node needs an exact Ollama model name, and model names like qwen2.5vl:7b-instruct-q8_0 are the kind of string you will mistype, misremember, or get slightly wrong after you pull a different quant. This node fetches the real list from your server so you can't.

Here's what it does, and it's a one-trick pony done well. When the node is created, and again whenever you press its Fetch button, it requests GET /api/tags from the Ollama server and populates an available_models dropdown. Pick an entry and it copies the exact name into the model field, which stays editable - so you can still type a name the server doesn't currently report, for example one you're about to pull. That's the whole job: two string outputs, URL and model, that plug straight into the Generate node's inputs.

One detail worth knowing: model discovery is proxied through the ComfyUI server rather than called from the browser, which sidesteps CORS entirely. That's the classic wall people hit when they try to query a local Ollama from the web UI, and it's the reason this node works on a fresh setup instead of after an afternoon of fiddling.

The three inputs mirror the three widgets: url (default http://127.0.0.1:11434, and only http/https are accepted), available_models (the fetched list, read-only-ish), and model (the editable name that becomes the output). There are no optional inputs and no secrets - no API key anywhere in this pack.

If the Fetch comes back empty or errors, the usual culprits are: Ollama isn't running (ollama serve if you installed it as a bare binary), the URL points at the wrong host or port, or you've included a query string or fragment (this node rejects those). A non-loopback URL is fine and will work - the proxy handles it - just remember the privacy note from the pack README: everything you send gets transmitted to whatever server you configure.

Install is the same for every node in this pack:

cd ComfyUI/custom_nodes
git clone https://github.com/craftingmod/ComfyUI-Ollama-ImageList

Then restart ComfyUI, or install "Ollama Image List" from ComfyUI Manager. You'll also want an Ollama server and a vision model pulled before the Generate node has anything useful to say - this node just finds the names.

Honestly, you could skip this node and type the model name by hand. But "I changed the quant and forgot to update the workflow" is a failure mode I've hit more than once, and a dropdown fed from the live server removes the whole class of typos. It's the kind of utility node that earns its place the first time you press Fetch after pulling a new model.