☁️BizyAir NanoBananaPro

What it is

Nano Banana Pro is Google's flagship image model - officially Gemini 3 Pro Image, informally named after an anonymous codename that leaked on a HuggingFace leaderboard and then got embraced, banana emojis and all, by Google's own execs. It does 4K native output, takes up to 14 reference images at once, and runs a "thinking mode" that reasons about your prompt before generating. This node is BizyAir's ComfyUI wrapper around that Google API - you're not running anything locally, you're paying Google's per-image rate (Google lists $0.039–$0.24 depending on resolution) through your BizyAir account.

You'd reach for it when you want Google's editing and multi-reference capabilities inside a ComfyUI graph instead of jumping out to ImageFX - character consistency across edits, precise text rendering, or combining several reference images into one composite.

How it works

You write a prompt, optionally attach one or more reference images, and pick an operation. The node sends all of that to Gemini 3 Pro Image and returns whatever comes back - an edited or generated image, plus a text response from the model (Gemini's "thinking" commentary, most likely, though the node doesn't label it further than string).

Inputs and outputs that matter

prompt is required; operation picks the mode - generate, edit, style_transfer, or object_insertion - which is the main lever for what the node actually does with your reference images. aspect_ratio covers the usual spread (1:1 through 21:9, plus auto), and resolution steps from 1K to 4K or auto. temperature and top_p are standard generation-sampling knobs if you want more or less variation; leave them at 1 unless you have a specific reason not to. seed and max_tokens round out the required fields.

On the optional side, images is where you attach your reference photo(s); inputcount (1–14) dynamically controls how many reference-image slots the node exposes - bump it up and click "Update inputs" to add more sockets, which is exactly the multi-reference capability the KB flags as Nano Banana Pro's standout feature. quality (standard/high) and character_consistency (on by default) round out the optional inputs - the latter is worth knowing about specifically if you're doing sequential edits of the same subject and outputs keep drifting.

Outputs: image is your result, string is the model's accompanying text, and bizyair_model_name is an internal routing tag for logging, not something you'd normally wire anywhere.

Installing it

Install via ComfyUI Manager (search "BizyAir") or git clone https://github.com/siliconflow/BizyAir.git into custom_nodes, then restart. A BizyAir API key is mandatory before any node in the pack runs - register through the login prompt on first launch.

Common issues

Because Google's own safety filtering sits in front of this model, expect refusals or heavily sanitized output on anything remotely risky - that's Google's policy, not a BizyAir bug, and there's no way to route around it from this node. If your multi-reference generations aren't picking up all your images, double-check inputcount actually matches the number of sockets you've filled - it's a manual slider, not automatic. And since this bills per call at Google's rates, a workflow that loops this node in a batch can get expensive fast; keep an eye on your BizyAir credit balance if you're iterating.