Get Gemini Models JNK

Get Gemini Models is a pure configuration node with a deceptively simple job: it holds your Gemini model IDs and their RPM (requests-per-minute) limits and hands them out as strings and ints. No API calls, no network, no loading. Open the source and you'll find a function that literally returns its inputs unchanged. It's the "settings panel" of the JNK pack's Gemini trio - the other two being Get Gemini Keys (keys from a file) and Ask Google Gemini (the actual API call).

Why bother with a node that does nothing? Because it gives you one place to change your model lineup instead of editing a dozen Ask Google Gemini nodes. Say you've built a batch workflow that captions, expands, and critiques images, each with its own Ask node. Bump from gemini-2.5-flash to a newer model and you'd normally have to find every node. Wire each Ask node's model input to this node's outputs instead, and you update the whole graph by editing one widget.

Inputs and outputs

All inputs are optional, and each mirrors to an output of the same name:

• pro (default gemini-2.5-pro) → pro, with pro_rpm (default 5) → pro_rpm
• flash (default gemini-2.5-flash) → flash, with flash_rpm (default 10) → flash_rpm
• light (default empty) → light, with light_rpm (default 15) → light_rpm

The pattern: model ID string plus its RPM int, three tiers. The light slot exists because Google has shipped lightweight flash-lite tiers that change names frequently - leave it empty until you know the current ID.

Installing it

Pack install: search JNK in ComfyUI Manager, or

cd ComfyUI/custom_nodes/
git clone https://github.com/Aljnk/ComfyUI-JNK-Tiny-Nodes.git

No dependencies for this node beyond core.

Where people get burned

The defaults encode a reasonable free-tier assumption - pro is slow and rare (5 RPM), flash is the workhorse (10), light is fastest (15) - but they're guesses, not gospel. Check the current rate limits because free-tier numbers have shifted, and an RPM set higher than the API allows will just earn you 429s that throttle harder than waiting would have.

The other gotcha: this node validates nothing. It will happily hand out a typo'd model ID, and the failure only surfaces later as an error from Ask Google Gemini - never here. Model IDs on the Gemini API have churned a lot (the 2.5 line, flash-lite, the Nano Banana rebrands), so when things break after an update, check whether your string still matches a real model before debugging anything else. It's config-in-a-box, and config-in-a-box only protects you from editing fatigue, not from typing mistakes.