DataSet_GroqChat

Groq is the API that made "fast LLM inference" feel like a solved problem - their hardware runs models like Llama at ridiculous speeds, and their free tier is famously generous. DataSet_GroqChat wraps that into a ComfyUI node, so you can generate prompts, normalize captions, or rewrite text from inside a graph without paying OpenAI prices.

It's the value pick of the pack's three chat nodes. If you want to experiment with LLM-powered captioning or prompt generation and you're not already spending money on OpenAI or Anthropic, this is the one to start with.

The dropdown that lies to you

Here's the gotcha that trips up basically everyone on first use: the model dropdown is populated at import time by calling the Groq API for your available models. If GROQ_API_KEY isn't set in your environment - or the API call fails - the dropdown contains exactly one entry: Key Invalid. It's not an error message, it's a model option. If you see Key Invalid in the list, your key is missing or wrong.

Fix it before launching ComfyUI:

export GROQ_API_KEY=gsk_...

Then restart. With a working key, the dropdown fills with the real model IDs Groq serves (the Llama and other open models), and the same restart picks up your system_prompt and user_prompt fields.

Inputs and output

• model - chosen from the live list, once the key is set.
• system_prompt - the persona/instructions, multiline.
• user_prompt - the actual request, multiline.
• max_tokens - response cap, default 1024.

Output is a single STRING. The node runs at a fixed temperature of 0.5 - you can't change it, but it's a sane middle ground for captioning and prompt work.

How it works

It uses the official groq Python client against https://api.groq.com/openai/v1 - the endpoint is OpenAI-compatible, which is why the SDK feels familiar. Every run is a live API call: no local model, no VRAM, just a network round trip. Because Groq is fast, that round trip is usually a couple of seconds rather than the tens of seconds you'd wait on other providers.

Installing

It's part of the ComfyUI-DataSet pack:

cd ComfyUI/custom_nodes
git clone https://github.com/daxcay/ComfyUI-DataSet.git
cd ComfyUI-DataSet && pip install -r requirements.txt

Restart after. The groq client comes from requirements.txt. And remember the pack-wide convention: errors come back as STRINGs starting with Error: rather than failing the queue, so an empty or Error: output almost always means a missing key or a network problem.

The author (Daxton Caylor) keeps this whole pack on env-var keys by design - it was a deliberate change in the Nov 2024 update. Set the variable once, restart, and the node just works. That's the whole setup.