gemini / 2.5-flash / createImage
Gemini 2.5 Flash text-to-image, a.k.a. the base Nano Banana, as a Comfy node
- api_config
- images
- errors
- workflow_id
- raw_json
Here's a fun bit of naming archaeology: "Nano Banana" was the community codename that became official Google branding for Gemini image generation - and the base Nano Banana is Gemini 2.5 Flash Image. So this node, CivitaiImageGenGemini25FlashCreateImage, is literally the classic Nano Banana inside a ComfyUI node, hosted on Civitai's cloud. Which is great, because Google's own API setup is a trip through Gemini API or Vertex AI, billing accounts and all; this node is the shortcut that skips it.
It's part of the official civitai/civitai-comfy-nodes pack (engine gemini, model 2.5-flash, operation createImage). Type a prompt, it generates on Civitai's fleet, and the result drops into your graph as an IMAGE. No local model, no key at Google, no VRAM. Your cost is Buzz.
How it works. The machine behind every node in this pack: submit a workflow payload to orchestration.civitai.com, long-poll until it finishes, convert the output. This node's input surface is deliberately tiny - in fact it's the sparsest create node in the image category. prompt is the only required field, and the optional list is short: quantity (1-4), output_format (jpeg/png/webP), image_metadata (external metadata stored with the image), and api_config. No width/height, no seed, no negative prompt, no sampler. Google's model decides the canvas, and Gemini's image models are notoriously picky about their own parameters - that's why the pack doesn't expose them.
Inputs that matter. Honestly, just prompt. Gemini 2.5 Flash is an instruction-following model that rewards full sentences over tag soup - describe the scene, the style, the lighting. quantity is there if you want a few candidates at once. Everything else is optional and defaults to sane values. If you want the dials - resolution, aspect ratio, reference images, web search - you want the nano-banana-2 or nano-banana-pro nodes instead, which expose those.
Install. With the whole pack, since these are generated nodes that ship together:
cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt # requests
Restart, and look under the Civitai/Image/gemini menu. Or use ComfyUI Manager → Custom Nodes Manager → search Civitai Comfy Nodes → Install.
Auth and gotchas. Credentials resolve from a wired Civitai Auth node, then CIVITAI_API_TOKEN, then a stored key or OAuth from the Civitai sidebar. The pack's error hints are the ones to know: 402 = insufficient Buzz, 401 = token missing/expired, 429 = rate limited. Jobs time out at 30 minutes, and early-preview means the node's shape can change between releases.
The wider caveat applies in full here: Gemini image models are among the most heavily censored you can call, and Civitai enforces the same restrictions as its web generator. Your prompt leaves your machine and goes through two layers of filtering. If the content clears, though, you've got the base Nano Banana in your graph with the least setup Google has ever offered - and for that, the tradeoff is often worth it.
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| output_formatopt | COMBO | 4 options: , jpeg, png, webP | |
| image_metadataopt | STRING | External metadata that will be stored with the image | |
| quantityopt | INT | 11–4 | — |
| api_configopt | CIVITAI_CONFIG | Optional Civitai Auth connection; defaults to CIVITAI_API_TOKEN or stored OAuth login. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | — |
| errors | STRING | — |
| workflow_id | STRING | — |
| raw_json | STRING | — |