🔱 Gemini Question

Gemini Question is the pack's general-purpose Q&A node: you type a question, it asks Gemini, and you get the answer back as a string. It's the simplest node in the Artha-Gemini family, and also the one with the clearest limit - the author says it outright in the node's own description: the context is not preserved, and this node is not suitable for dialogue.

That's a feature, not a bug. In a ComfyUI pipeline, stateless one-shot Q&A is usually exactly what you want. You don't need a chatbot with memory; you need a node that can answer "what's the color temperature of golden hour in Kelvin?" and give you a string to use elsewhere. Question is that node.

How it works

Nothing fancy. The node takes your question, wraps it in a minimal system instruction (You are an intelligent ai assistant.), and sends it to the selected Gemini model in a single call. The response comes back as plain text. Because there's no conversation history, each run is fully independent and reproducible given the same inputs - which is what makes it usable inside a deterministic workflow at all.

It's the base node of the pack in the literal sense: several other Gemini nodes (Operation, Translate, Condense) are structured exactly like it, just with their own system instructions and extra inputs. Question is the unadorned version.

Inputs and outputs

• question - the text you're asking. Multiline, no default.
• api_key, model, max_tokens, temperature - the shared API block. Default model gemini-2.5-flash; temperature defaults to 0.7.

Output: response (STRING). Wire it into a text display (or the pack's Gemini Response node) to see the answer.

Installing it

Via ComfyUI Manager (search Artha-Gemini) or:

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

No model downloads; dependencies are google-genai, Pillow, numpy, torch, and soundfile. You need a Gemini API key from Google AI Studio - this node always makes a live call.

Where people get burned

The big one is expecting conversation. Ask it a follow-up question and it has no idea what you asked first - that's by design, and if you need multi-turn dialogue, this pack isn't the tool. The workaround, if you genuinely need context, is to include the prior context in the question text yourself, or chain Question nodes.

Second, temperature matters more here than you'd think. At the default 0.7, factual answers come back with noticeable variation between runs. If you're using Question for reference lookups (an API spec, a parameter default), drop temperature toward 0 for more consistent answers.

The usual pack gotchas: keep the key in api.json (gemini_api_key) or GEMINI_API_KEY env var, never the plain-text workflow field; remember it's a billable call every run; and when the API fails you'll get a response starting with Error: with the real message in the console. For a question node, that's the most useful diagnostic in the pack - read it.