Griptape Prompt Driver: Google

The Google prompt driver is the budget pick of the hosted brains in this pack, and it punches above its price. Gemini 2.0 Flash - the default model - is genuinely fast, genuinely cheap, and genuinely multimodal: it can describe images, which matters a lot in a tool built on top of an image-generation app. If you're experimenting with LLM agents in ComfyUI and don't want to commit a big API budget, this is a great first driver to wire in. The free tier of the Gemini API is surprisingly usable for hobby projects.

It works like every other driver in the pack: the node constructs a Griptape GooglePromptDriver with your model and sampling settings, and emits it on DRIVER for you to plug into an agent configuration or RAG module.

The inputs that matter

• model - dropdown of current Gemini models. Default gemini-2.0-flash; gemini-2.0-flash-lite-preview-02-05 is the cheapest tier, gemini-1.5-pro the heavyweight. Flash is the right default - pro only earns its cost on long reasoning tasks.
• google_api_key_env_var - env-var name, default GOOGLE_API_KEY. Keys come from AI Studio (makersuite.google.com/app/apikey), and the free tier is genuinely free for light use.
• top_k (default 250) and top_p (default 0.999) - Gemini's native sampling controls, both reasonably high so output stays close to the model's best guess.
• max_tokens - -1 default; raise it for long outputs.

Standard temperature, seed, use_native_tools, and max_attempts_on_fail are all there. Output: DRIVER.

Why you'd actually pick it

Beyond price, Gemini's multimodal strength is the sleeper feature here. The pack's image-description and image-query flows work with several models, but Gemini has the most forgiving vision support among the cheap options - feed it a busy reference image and it reliably tells you what's there. Pair that with a prompt driver and you've got a pipeline that looks at an image and reasons about it, which is a genuinely neat ComfyUI trick.

Installing it

Pack install, same as always: ComfyUI Manager → Griptape → ComfyUI-Griptape, restart, or

cd ComfyUI/custom_nodes
git clone https://github.com/griptape-ai/ComfyUI-Griptape

Then set GOOGLE_API_KEY in Settings → Griptape or your environment. Existing environment values show up automatically in the settings page.

Common issues

Env-var-name trap, again: the field wants GOOGLE_API_KEY, the name, not the key itself. Also be careful with the older gemini-1.0-pro / gemini-pro dropdown entries - legacy models get retired by Google with short notice, and a deprecated model fails with a "model not found" style error. If that happens, bump to a current flash or pro version.

Pack-wide: if ComfyUI breaks after install, it's griptape[all] clobbering torch - reinstall with pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121. Stale griptape shows up as import errors; python -m pip install griptape -U sorts it.