LLM Chat

This is the node where the art-venture LLM chain finally does something. The API node holds your key, the Config node holds the settings, the Message node builds the conversation - and this one takes all three, sends the request, and gives you back the model's reply. It's the hub that ties the pattern together, and it returns both text and, for image-capable models, an image.

That second output is the interesting part. Because the pack routes image models (Gemini's NanoBanana) through the same chat interface, LLM Chat is how you'd generate an image from an LLM inside ComfyUI - describe what you want in a message, and the picture comes out the images output.

How it works

You give it three things: the messages (your conversation, from LLM Message), the api (a connection handle from Gemini API, Bedrock Claude, etc.), and the config (model and settings from LLM API Config or NanoBanana API Config). It runs the call and returns the result. That's the entire three-legs-plus-runner shape: credentials, settings, conversation, execute.

The inputs and outputs

• messages (LLM_MESSAGE) - the conversation to send.
• api (LLM_API) - the provider connection.
• config (LLM_CONFIG) - model and generation settings.
• seed (INT) - for reproducibility. Fair warning: LLMs aren't as deterministic as diffusion samplers, so the same seed won't always give a byte-identical reply, especially at higher temperature.
• text (output, STRING) - the model's text response. Wire this into a prompt, a text overlay, a save-text node, or anywhere downstream.
• (optional) images (output, IMAGE) - the generated image, when you're running an image model.

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.

Common issues

Missing-node red means the pack isn't installed - Manager → "Install Missing Custom Nodes."

Most failures here aren't in this node, they're in what feeds it. All three inputs are required - no api, no config, or no messages and it can't run. If it errors, the usual causes are upstream: a bad or unauthorized key on the API node, a model your credentials can't reach, a stale model ID in the Config, or (on Bedrock) model access not enabled for your region. Because this is the node that makes the network call, it's where the error surfaces even though the cause is one node back - read the message, then check the API and Config nodes.

Expecting an image and getting nothing on the images output? Two checks: is your Config an image config (NanoBanana) rather than a text one, and does the model/modalities setting actually return an image. Point a text model at this and the image output stays empty by design. And since this is a live cloud call, it needs network access and will cost real API tokens per run - worth knowing before you drop it inside a big batch.