LLM Model Selector

LLM Model Selector is the node that stops you from memorizing model names. Wire it a client and it gives you a dropdown of every model currently sitting on your Ollama or llama.cpp server - pick one, and the model name comes out the other side, ready to feed straight into Chat Completion.

What it actually does

Here's the part that surprises people: it doesn't call the server at runtime. It's a pass-through node - a client in, a model name chosen from a dropdown, and the chosen string out. The source is explicit that it "never calls the provider." The dropdown is populated once, when ComfyUI starts, from the server you've connected to. Its whole job is being a friendly typed picker so your graph carries a model name around without you hardcoding strings.

That sounds trivial until you remember the alternative: typing qwen2.5:7b-instruct-q4_K_M into a text field on every chat node, and hoping you didn't typo it. A dropdown that's guaranteed to match what the server actually has is a real quality-of-life win, especially when you're swapping models to compare outputs.

Inputs and output

• client - required, LLM_CLIENT socket from Ollama Client or LlamaCpp Client.
• model - the dropdown. If it shows "(start Ollama - click ⟳ Refresh)", your server isn't reachable or wasn't running when ComfyUI booted.
• Output: model_name - a plain STRING you wire into Chat Completion's model input (or LLM Load Model's).

Install

Via ComfyUI Manager (search "comfydv"), or:

cd ComfyUI/custom_nodes
git clone https://github.com/darth-veitcher/comfydv.git

Restart ComfyUI and bring a backend: ollama serve (plus ollama pull qwen2.5:latest to grab a model) or llama.cpp's llama-server --models-dir ./models -c 8192.

Gotchas

• The dropdown populates once, at server start. Start Ollama before ComfyUI, or click the ⟳ Refresh button after pulling a new model. This node deliberately bypasses ComfyUI's frozen-at-startup combo validation, so a model pulled after boot won't fail with a spurious "value not available" error - but you still need that refresh to see it in the list.
• Old workflows call it OllamaModelSelector. If a saved graph reports a missing node, that's the rename - delete and re-add, the behavior is identical.
• It's a picker, not a loader. Selector only hands you the name. If you want the model pinned in memory before chat runs, put LLM Load Model in front instead - or just wire the selector's output straight into Chat Completion and let the server load lazily. For most casual use, that's plenty.