Gemini Image🧸

Gemini Image wires Google's Gemini Flash models into ComfyUI, and the thing that makes it different from the other image nodes in Plush is that it's genuinely multimodal in both directions. You can feed it text and an image, and it can return text and an image from a single prompt. So it's not just a text-to-image call, it's more like a small conversational image tool: describe an edit, get back the edited picture plus the model's commentary.

It's a paid closed API (Google's), and at the time this node shipped these Gemini image models were new and, frankly, twitchy. Useful, but expect rough edges.

How it works

You give it a prompt (optionally with an input image), it calls the Gemini API, and you get back both an IMAGE and a text output. Passing an image in lets Gemini use it as reference or as something to edit; leaving it out is plain text-to-image. Because Gemini returns text alongside the image, the text output is where you'll often find the model explaining what it did, or, importantly, why it refused.

The inputs that matter

• prompt (multiline): your instruction. Multimodal models respond well to plain, direct language.
• image (optional): a reference or edit target.
• creative_latitude (0.1 to 1.9, default 0.7): the temperature knob.
• tokens (default 800): output cap.
• Custom_ApiKey (optional): if your key lives in a custom-named environment variable, feed it via the Custom API Key node. Otherwise the node uses the standard GEMINI_API_KEY variable.
• Add_Parameter (optional): inject extra API parameters via the Add Parameters node.

Outputs: image, text, plus help and troubleshooting.

The safety-filter reality

Here's the thing that will actually cost you time. Google's Gemini image models ship with strict safety filters, and the author calls this out directly in the pack. Words that could connote children (young, child, boy, girl, baby) will frequently trip a safety refusal even in completely innocent prompts. When that happens the generation fails and the reason shows up in the troubleshooting output. This isn't the node misbehaving; it's Google's policy layer. The author's own note is that these models were new and the provider tends to relax filters as they mature, so it's partly a matter of timing. For now, if a benign prompt keeps failing, look for a trigger word and rephrase.

Installing it

Grab Plush-for-ComfyUI through ComfyUI Manager (search "Plush", install Plush-for-ComfyUI, restart), or clone it: cd ComfyUI/custom_nodes && git clone https://github.com/glibsonoran/Plush-for-ComfyUI.git, then pip install -r requirements.txt in the folder, restart.

Common issues

You need a Google Gemini API key in the GEMINI_API_KEY environment variable (or a custom variable read through the Custom API Key node), and env vars generally need a machine reboot before they register. After that, the number-one failure is the safety filter, so read the troubleshooting output first, it usually names the reason. Because it's a paid API, set a spending limit on the key. And keep in mind the model list here is small and tied to whatever Gemini Flash version the node targets, so this is a "use it for the multimodal trick" node rather than a general workhorse.