Nano Banana Pro (OpenRouter Gemini Image)

"Nano Banana" is Google's own branding for its Gemini-native image generation - the name started as an anonymous leaderboard codename, Google execs started dropping banana emojis about it on social media, and it stuck hard enough to become the official product name. Nano Banana Pro specifically is Gemini 3 Pro Image: 4K native output, up to 14 reference images, and a "thinking mode" that reasons about the prompt before generating. This node gets you that model inside ComfyUI, but routed through OpenRouter rather than talking to Google directly - meaning you authenticate with an OpenRouter API key, not a Google one.

Why go through an aggregator instead of Google's own API? Mainly convenience if you already use OpenRouter for other models: one key, one billing account, and this node's custom_model field lets you point it at any other OpenRouter-hosted image model without touching the node's code. It's real, if modest, usage pattern in the ComfyUI world - OpenRouter shows up in community discussion at a steady trickle rather than a flood, more of a "power user already has this key set up" tool than something beginners reach for first.

What you actually configure

• prompt - your image description.
• model - a dropdown of four OpenRouter-hosted options, including google/gemini-3-pro-image-preview:nitro (this is Nano Banana Pro) and a ByteDance alternative, bytedance-seed/seedream-4.5. The :nitro suffix is OpenRouter's own tag for a throughput-optimized routing variant of that model - you get the same model, just routed for speed over cost.
• custom_model (optional) - overrides the dropdown entirely if you want to point at some other OpenRouter image model by its slug.
• aspect_ratio and resolution (1K/2K/4K) - straightforward generation controls; 4K is the headline feature of the Pro tier specifically.
• response_modalities - IMAGE+TEXT or IMAGE alone, depending on whether you want the model's accompanying text output or just the picture.
• images (optional, IMAGE) - reference images for image-to-image or multi-reference generation.
• files (optional, GEMINI_INPUT_FILES) - additional file inputs beyond plain images.
• system_prompt - editable, but ships with a strong default that instructs the model to always produce an image and to interpret any input as a literal visual directive rather than a conversation. Leave it unless you have a specific reason to change how the model interprets ambiguous prompts.
• openrouter_key - your OpenRouter API key, required.

Outputs: image and text - the generated image plus whatever text the model returned alongside it (empty if you set response_modalities to image-only).

Installing it

Search ComfyUI-load-lora-from-url in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/bollerdominik/ComfyUI-load-lora-from-url

No model to download here - this is an API node. You do need an OpenRouter account with credit on it and a generated API key; there's no free tier to speak of once you're actually generating images.

Where people get burned

• Don't expect it to do anything mature. Nano Banana is community-described as "powerful, unimaginative, massive hidden biasing" - it's a genuinely capable model that's also aggressively filtered, and that filtering happens on Google's side regardless of which route you access it through. If your use case brushes up against NSFW or edgy content at all, this isn't the model for it.
• The key field is a plaintext string field in the node, not a ComfyUI credential manager entry. If you're sharing this workflow with anyone, remember to strip your key out first.
• 4K resolution costs meaningfully more per image than 1K on Google's own pricing for this model (roughly $0.04 to $0.24 depending on resolution) - OpenRouter's pass-through pricing will track that. Don't default to 4K while you're still iterating on a prompt; drop to 1K until you're happy with the composition, then bump the resolution for the final render.