LM Studio Generate V2

Generate V2 is the node the pack's V2 design is actually organized around. Everything else in the "V2" family exists to feed this one: Connectivity V2 gives it the URL and model, Options V2 gives it the sampling settings, and Generate V2 itself does the request. It's also the most capable node in the pack, because it quietly does vision too - plug an image into its images input and it'll behave like LM Studio Vision on top of everything else. If you're starting fresh with this pack, this is the node to build your workflow around.

Why you'd reach for it

Modularity is the pitch. The original nodes in this pack each carried their own URL, model, and sampler settings, which meant retyping everything for every node and hunting through a big graph to change one thing. V2 splits that apart, and Generate V2 is the consumer: you wire one Connectivity V2 and one Options V2 into it, and change the model or the temperature in exactly one place. That's the workflow-design win - shared config, replaceable mid-graph. The images input makes it a text-and-vision node in one, which is the fastest way to build the image-to-prompt-to-diffusion loop without a separate Vision node.

How it works

Like the rest of the pack it posts to {url}/v1/chat/completions on LM Studio - but it only learns the URL and model from its connectivity input, and it only sends the sampler settings whose enable_ flags are on in its options input. Your system and prompt are the only required fields; if an images input is present, the images are base64-encoded as PNG data URLs and attached to the user turn (so it needs a vision-capable model loaded). filter_thinking strips reasoning tags by default.

The novel piece is meta. Generate V2 emits an LMSTUDIO_META output that carries the connectivity and options it just used, and it can accept one on the way in. Chain V2 nodes together and the next one can reuse the previous node's connection and settings without you wiring Connectivity and Options again - or you can override them by wiring fresh cards. It's the pack's way of letting configuration flow through a graph, and it's why this node ends up with four outputs.

The inputs

• system and prompt - the only required ones.
• connectivity - LMSTUDIO_CONNECTIVITY from Connectivity V2. Required at runtime: the node raises an error if neither this nor meta supplies it.
• options - LMSTUDIO_OPTIONS from Options V2. Optional; without it, the model gets defaults.
• images - optional IMAGE input for vision.
• previous_conv - a conversation thread to continue.
• meta - an LMSTUDIO_META from a previous Generate V2.

The outputs

• result - the model's reply. This is the one you wire into your text encoder.
• previous_conv - conversation as JSON.
• meta - the connectivity + options bundle for the next V2 node.
• read_conv - the conversation formatted for humans.

Installing

Standard pack install: ComfyUI Manager → search "comfyui-lmstudio", or

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

restart ComfyUI, load a model in LM Studio, start its local server on port 1234. No model downloads through the pack. If ComfyUI errors on a missing httpx, pip install httpx and restart.

Where people get burned

The error message "Required input 'connectivity' or 'connectivity' in 'meta'" is this node's way of telling you you forgot the Connectivity V2 card - it's the single most common stumble with the V2 family. Second: feed it an image while a text-only model is loaded and you'll get a runtime error, not a helpful hint, so match the model to the inputs. And keep the VRAM reality in mind - the LLM and the diffusion model share the card, so leave the enhancement model small and unload it from LM Studio while you sample.