NS LLM Model Selector

NS LLM Model Selector is the "central model picker" for the pack's LLM nodes - the same 11-model roster as NS LLM Chat (Claude, Gemini, GPT, Grok), but instead of calling anything it just emits the model name as a STRING. You wire that string into any number of NS LLM Chat nodes' model_override inputs, and one dropdown controls them all.

It exists because of a real workflow pain: once your graph has three LLM steps - say a prompt enhancer, a captioner, and a summarizer - "the model is selected in three different places" becomes a live bug generator. You're testing with gemini-3.5-flash because it's cheap and fast, you forget one node is still on claude-opus-5, and suddenly one leg of your pipeline is five times slower (and pricier) than the others. This node is the "one knob" fix: flip it and every wired chat node follows.

How it works

Input is one model dropdown (default claude-opus-5), output is the chosen name as a model STRING. That's the whole node. The mechanism is the shared model list in the pack - the same all_models list that populates NS LLM Chat's dropdown - so the selector and the chat nodes can never disagree about what names are valid. It doesn't touch keys, doesn't call anything, does no work at runtime beyond returning the string.

How to wire it

Connect the model output to the model_override input on each NS LLM Chat node you want to control. When wired, the override wins over the chat node's own dropdown - which is exactly what you want from a central picker. Leave a chat node's override disconnected and it just uses its own widget as before.

Install

Part of symbiotica-ai/comfyui-nodes - ComfyUI Manager search "Symbiotica", or clone + pip install -r requirements.txt. No API key, no ffmpeg, no model download - it's a pure string pass-through. Restart ComfyUI after installing.

Gotchas

Two things to keep straight. First, model_override is a STRING input, so if you connect the wrong output type it won't seat - that's the socket telling you it's not the right wire. Second, the selector only affects the pack's NS LLM Chat nodes; it won't steer the pack's dedicated Gemini/Claude/Grok image nodes, which have their own model enums - don't wire it there expecting control. And because it's a shared roster, an older workflow saved with a model name that's since been renamed will error at load rather than silently picking the closest thing - refresh the dropdown and re-pick.