LM Studio Options V2

LM Studio Options V2 is a settings card, not a generator. It collects a handful of model parameters - temperature, top-p, seed, max tokens - and hands them to a generation node as a single LMSTUDIO_OPTIONS bundle. What makes it interesting is the pattern of enable_ toggles: each parameter has a matching boolean, and only the parameters you tick actually get sent to the model. It's a deliberate "send exactly what I asked for" design, not a bug.

Why you'd reach for it

It's the second half of the V2 split in this pack. LM Studio Connectivity V2 holds the URL and model; this node holds the sampling settings; LM Studio Generate V2 does the actual work. The value of a separate options node is the same as a separate connectivity node: one card feeding several generation nodes. Build an options card once with your tuned settings, wire it into every Generate V2, and you can change the whole workflow's behavior from a single place instead of hunting through each node.

How it works

Under the hood the node just packages its inputs into a dict and returns it - no network call. The clever part is in the consumer, LM Studio Generate V2: it walks the dict and only forwards parameters whose enable_ flag is true. So if enable_seed is off, no seed is sent and LM Studio rolls its own; if it's on, your seed goes along. max_tokens follows the same convention the pack uses everywhere else: -1 means no cap, and anything above that is the limit. debug dumps the whole options dict to the console so you can verify what's actually enabled.

The inputs

• enable_temperature / temperature - tick the box, set the value (0.8 default).
• enable_top_p / top_p - nucleus sampling threshold, 0.9 default.
• enable_seed / seed - tick to make runs reproducible across the chain.
• enable_max_tokens / max_tokens - tick and set a cap (max 4096) if a model rambles.
• debug - print the assembled options to the console.

That's a genuine short list: for most workflows you'll tick temperature, maybe seed, and nothing else.

Outputs

• options - the LMSTUDIO_OPTIONS bundle. This is the output you use; it plugs into the options input of LM Studio Generate V2.
• previous_conv and read_conv - conversation passthrough, present for symmetry with the rest of the pack. Wire them along if you're threading a chat.

Installing

The standard pack install: 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; if ComfyUI errors on httpx, pip install httpx and restart. Nothing here calls LM Studio directly, so this node works even while the LM Studio server is down - the failure only shows up when a generation node actually makes a request.

The honest take

The enable-toggle design is genuinely tidy - it's the difference between "the server always overrides some of what you sent" and "the server gets exactly the knobs you turned." Just remember the settings do nothing until you tick their box; the temptation is to set a temperature, watch nothing change, and blame the node. If you tick debug and see the enabled set in the console, you'll know exactly what got sent.