Placeholder: Gemini 2.5 Flash Image (🧪)

The name tells you almost everything: this is a placeholder for Google's Gemini 2.5 Flash Image model - the one the community calls Nano Banana 🍌. It generates a dummy image locally with Pillow, calls no API, needs no API key, and costs zero. You drop it into a workflow where a real Nano Banana node will eventually live, and it quietly draws a dark-gray box with a gold border, a crosshatch pattern, and every parameter you passed rendered as text on the image.

Why does that exist? Because the real Nano Banana charges you per image and rate-limits you, and nothing kills a prototyping session faster than paying for forty test generations while you fiddle with wiring. This node lets you build and stress the structure of a workflow - branches, compositing, resizes, loops - without touching the API. The README's framing is exactly right: verify output image sizes, check connections, iterate on the graph, then swap in the real node when you're ready to spend.

How it works

Under the hood it's pure Pillow, which is why it's instant and free. The node reads your aspect_ratio and maps it to actual pixel dimensions using a 1024px long edge - so 16:9 becomes 1024×576, 1:1 becomes 1024×1024, and dimensions get rounded to a multiple of 8. Then it draws the background, border, crosshatch, and a text overlay listing your prompt and all the sampling parameters, and converts the result to a standard ComfyUI IMAGE tensor.

The clever bit for prototyping: the parameters are rendered on the image itself, so you can glance at the preview and confirm the node actually received what you think you set. Your seed, temperature, top_p, and top_k are drawn onto the image but have zero effect on the output - they're there purely so the node mirrors the real Nano Banana node's inputs and the swap-in-place works.

Inputs and outputs that matter

The node mirrors the Gemini 2.5 Flash Image node's inputs, so you can wire them identically:

• aspect_ratio - the one that actually changes the output. Options go from 1:1 through 21:9 plus an auto default. This is what you tune to verify downstream sizes.
• prompt - rendered onto the image (wrapped to three lines), handy as a visual label.
• output_mode (IMAGE+TEXT / IMAGE) - noted on the overlay; both outputs are always produced.
• model - locked to GEMINI_25_FLASH_IMAGE, just like the real node.
• Optional image1/image2/image3 inputs accept image tensors so workflow wires don't break, but the placeholder doesn't draw them - it's a dummy, after all.

Outputs are Placeholder Image (IMAGE, wire it to a preview or Save Image node) and param_summary (STRING, a key: value list of everything passed - handy for logging or a text display node).

Installing it

This is part of the catnipglitch/comfyui-genmedia-gemini-api pack, a community fork of Google's official comfyui-google-genmedia-custom-nodes. The easiest route is ComfyUI Manager → Custom Node Manager → search comfyui-google-genmedia-custom-nodes → Install. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/catnipglitch/comfyui-genmedia-gemini-api
pip install -r comfyui-genmedia-gemini-api/requirements.txt

Then restart ComfyUI. Heads up: the requirements pull in the whole Google stack - google-genai, google-cloud-aiplatform, moviepy, opencv-python-headless and friends - because the pack also ships the Veo and Lyria nodes. It's heavy, but you only install it once.

Common issues

• It errors about GCP / API keys. Not this node's fault. The placeholder makes no API call at all - if you're seeing auth errors, they're coming from a real generation node elsewhere in your pack, or the placeholder was swapped in a workflow that still routes something into an API node.
• The image looks "wrong" for your aspect ratio. Remember the long edge is fixed at 1024 for the 2.5 placeholder - that's intentional so it mirrors the real node's output size. If you're testing 2K/4K canvases, you want the 3.1 or 3 Pro placeholders instead.

One last tip: since the real Nano Banana node lives in the same pack, you can build with the placeholder, then switch node types and only your wallet notices.