🔱 Artha Gemini Instruct

Artha Gemini Instruct is the pack's note-taking node. It doesn't call any API, it doesn't need a key, and it doesn't do anything clever - it takes three text boxes (role, task, instructions), glues them into one formatted block, and hands it out as a single output. That's the whole job, and for the Artha system it's an important one, because the rest of the Gemini nodes lean hard on system-level instructions to steer the model.

The description says it "provides system-level instructions for other Gemini nodes," and that's the design intent: write your role and task once, keep them tidy in one place, and feed the assembled instruction into the pipeline. The output type is ARTHAINSTRUCT, the pack's own custom string type, so it's a structured value rather than a loose wire.

The three inputs

• role - defaults to "You are an intelligent AI assistant." This is who the model should be: a photographer, a portrait expert, a "master of conciseness."
• task - defaults to "Your task is to...". What the model should do.
• instructions - free-form extra guidance. Empty by default.

Execute it and the node assembles them with a bit of formatting - quotes stripped out, role and task on their own labeled lines, instructions appended - and returns one string. It's the same shape the pack's own built-in agent prompts use internally, so you can think of this as writing your own agent file without touching the filesystem.

Where it fits (and the honest caveat)

Here's the thing to know before you build around it: the ARTHAINSTRUCT output is a custom type, and the plain system_instruction inputs on nodes like Artha Gemini Prompter, Artha Gemini Motion, and Artha Gemini Imagen are ordinary STRING slots. Custom types don't snap into plain string inputs, so in practice you'll usually want to keep this node's output readable - route it through a display node or just copy the assembled text into the system_instruction field of whichever node you're steering. It's a builder and a scratchpad, not a magic wire.

That's fine, honestly. Its real value is workflow hygiene: instead of burying a long role prompt in a text widget you can't see, you keep the three pieces separate, edit them independently, and know exactly what the model was told when you're debugging why an output drifted. It's the kind of node that quietly earns its keep in bigger graphs.

The catch

There basically isn't one, which is refreshing for this pack. No key, no network, no failure mode beyond empty inputs producing an empty-ish block. The only trap is the same one that hits all of Artha: the pack needs to be installed and the node list refreshed before you'll see it, and it shares a requirements install with the rest of the pack.

Install

Part of the one-install 🔱 pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Cyrostar/ComfyUI-Artha-Nodes
cd ComfyUI-Artha-Nodes
pip install -r requirements.txt

Restart ComfyUI, or use ComfyUI Manager → search "Artha." Since this node never calls Google, you can use it even if you haven't set up a Gemini key yet - it's the pack's rare free lunch.