Placeholder: Gemini 3 Pro Image (🧪)

Of the three placeholders in this pack, this is the one you'll actually want to prototype with - because Gemini 3 Pro Image (Nano Banana Pro 🍌) is the expensive flagship, and its outputs are big. It generates native 4K images and is the one Google positions as the premium tier, with thinking mode and support for stacking multiple reference images. Every real 4K call costs real money, so it's the last node you want to be iterating on while your workflow structure is still wrong.

This placeholder mirrors the Pro node's interface exactly: it draws a dummy image locally with Pillow - dark gray background, gold border, crosshatch, all your parameters rendered as text - and it needs no API key, no GCP project, and makes zero network calls. You wire up your whole graph against it, confirm your downstream chain can swallow a 4096×4096 tensor, and only then swap in the real node.

How it works

Same local Pillow machinery as the other placeholders, with the Pro's parameter set. The image_size dropdown (1K / 2K / 4K) sets the long edge - 4K produces a 4096px canvas - and aspect_ratio (11 options, 1:1 through 21:9, plus auto) determines the other side, rounded to a multiple of 8. Everything else is drawn onto the image as an overlay and changes nothing. output_mime_type (PNG / JPEG) is mirrored and displayed so the swap-in works, though a dummy image is a dummy image in either format.

The one mirror that stands out: the Pro node accepts up to six reference images (image1 through image6) for multi-image editing, and this placeholder exposes all six input slots. They're accepted so your workflow wires don't break, but the placeholder doesn't render or use them - it's the aspect ratio and canvas size you're really testing here.

Inputs and outputs

• image_size - the whole reason this placeholder exists. Test 4K here to see if your VAE decode, upscaler, or compositing nodes survive a 4096px tensor before you pay for the real thing.
• aspect_ratio - sizes the canvas; auto defaults to square.
• prompt, temperature, top_p, top_k, output_mode - mirrored from the Pro node; rendered as text, behaviorally inert.
• output_mime_type - PNG / JPEG, informational.
• image1–image6 - optional IMAGE inputs, accepted but unused.

Outputs: Placeholder Image (IMAGE, wire to a preview or Save Image) and param_summary (STRING, a key: value dump of everything you passed - useful for logging or feeding a text node).

Installing it

Part of the catnipglitch/comfyui-genmedia-gemini-api pack, the community fork of Google's official comfyui-google-genmedia-custom-nodes. Via ComfyUI Manager, search comfyui-google-genmedia-custom-nodes and 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

Restart ComfyUI; the node appears under Google AI/Placeholder. No auth setup needed for the placeholder itself - that requirement only kicks in when you swap in the real Pro node and it starts calling the API.

Common issues

• "My 4K test image is a square at 1K." Check image_size - it defaults to 1K, and the placeholder's footer shows the actual pixel dimensions so you can confirm what you got.
• Auth errors in a placeholder workflow. Nothing here talks to Google. If you see project/API-key errors, a real generation node is still in your graph - find it and disconnect it while you prototype.
• Expecting the dummy to do anything with your six reference images. It won't - the input slots exist purely to keep the wiring identical. That's the deal with placeholders, and it's exactly why the swap-in works.