LM Studio Connectivity V2

Let's get the misleading name out of the way first: LM Studio Connectivity V2 makes no connection, sends no request, and pings nothing. It's a configuration card. What it does is bundle a URL and a model name into a single typed value - an LMSTUDIO_CONNECTIVITY - that the V2 generation nodes accept as one plug. If you've used the V2 family at all, this is how the graph learns where and what to ask.

Why you'd reach for it

The V2 redesign of this pack splits the old all-in-one nodes apart: connection settings, model settings, and the actual request each become their own node. Connectivity V2 is the "where do I point this and which model" half of that split. It earns its keep the moment you have more than one V2 node in a workflow - instead of retyping http://localhost:1234 and re-picking a model on every generation node, you wire one Connectivity V2 into several LM Studio Generate V2 nodes. Change the model once, and everything downstream follows.

How it works

Mechanically it's about as simple as a node gets: it returns a dict with your url and model in it, wrapped in the custom type. No network call happens at execution time. The model dropdown is still live though - like the rest of this pack, it's populated by a small endpoint the pack registers on ComfyUI's server, which asks LM Studio's /v1/models and fills the widget. If the dropdown is empty, the LM Studio server isn't running or isn't reachable at the URL you typed, not that anything is broken in this node.

The inputs

• url - where LM Studio's server lives. Default http://localhost:1234 is LM Studio's standard local-server port; change it only if you did.
• model - the model you want to talk to. It only lists what LM Studio reports as loaded, which is a handy sanity check in itself.
• previous_conv - optional passthrough, mostly there for symmetry with the rest of the pack.

The outputs

• connection - the LMSTUDIO_CONNECTIVITY bundle. This is the output you actually use: plug it into the connectivity input of LM Studio Generate V2.
• previous_conv and read_conv - the JSON conversation and its human-readable form, passed through untouched. Wire them along if you're building a multi-turn graph.

Installing

Same story as every node in this pack: ComfyUI Manager → search "comfyui-lmstudio", or

cd ComfyUI/custom_nodes
git clone https://github.com/SiegeKeebsOffical/comfyui-lmstudio

then restart ComfyUI. No models to download - they live in LM Studio. If ComfyUI complains about a missing httpx, pip install httpx and restart; the pack ships no requirements.txt so that one import isn't guaranteed.

The trap

The name implies a health check, so people expect this node to tell them when LM Studio is down. It won't - the failure surfaces later, as an exception from the generation node that actually makes the request. If you want to confirm connectivity, glance at the model dropdown: if it lists models, the server is up. And a note that applies to the whole pack: don't load a text-only model and expect vision requests to work - that's a runtime error you'll meet in the V2 generation node, not here.