LLM API Config

The art-venture pack lets you call real large language models - GPT, Claude, Gemini, Mistral - from inside a ComfyUI graph, and it splits that into three pieces: an API node (the connection and key), a Config node (the model and generation settings), and a Chat/Completion node (the thing that actually runs). This is the middle piece. It's where you pick which model to use, how many tokens it can spend, and how creative it's allowed to be.

The name trips people up, so let's clear it now: this node holds no API key and makes no connection. It's purely settings. You still need one of the API nodes - OpenAI, Gemini, Claude, Bedrock - to supply credentials. Config just describes how to run whatever model you're pointed at.

Why the three-node split

It sounds fussy but it's the right shape. Because config is separate, you can wire one settings node into several chat nodes, or swap the whole model roster by editing one box. It's the same "define once, reuse" logic as a sampler-settings node. For LLM-assisted prompting - where you might run the same model config across a few different prompts in one workflow - that separation pays off.

The inputs that matter

• model - a dropdown of 54 model IDs (defaulting to gpt-5.4-mini, and including the GPT-4o and GPT-4.1 families among many others). Pick the one your API node can actually reach.
• max_token - the output token budget (default 1024, up to 102400). Bigger allows longer replies; if a model gets cut off mid-sentence, this is usually the dial.
• temperature - randomness, 0 to 1 (default 0.5). Low for deterministic, factual, obey-the-instruction output; higher for looser, more varied text. Note the ceiling here is 1, not the 2 some APIs allow.
• custom_model (optional) - a free-text field for a model ID that isn't in the dropdown. This is your escape hatch when the roster is stale or you want a model the list doesn't name.

The single output is an LLM_CONFIG that wires into LLM Chat or LLM Completion.

Installing it

ComfyUI Manager → search comfyui-art-venture → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/sipherxyz/comfyui-art-venture

then restart ComfyUI. Pre-installed on comfy.icu. The LLM nodes lean on the respective provider SDKs, which the pack pulls in via its requirements on load.

Common issues

Red "missing" node in a shared workflow means art-venture isn't installed - Manager's "Install Missing Custom Nodes" fixes it.

The confusion that actually bites: you wire up this Config node, nothing else, and wonder why nothing runs. Config is one leg of a tripod - you also need an API node (with a valid key) and a Chat or Completion node. On its own it does nothing.

The other one is the model list going stale. This dropdown is a snapshot of whatever models the pack knew about when it was built; providers rename and retire models constantly. If the model you want isn't listed, or a listed one now errors, type the exact current model ID into custom_model rather than fighting the dropdown. And if replies keep truncating, raise max_token before assuming the model is broken.