Nano Banana Pro (Vertex)

Take the NanoBananaProNode writeup, keep every parameter, and swap the authentication for a Google Cloud service account - that's this node. NanoBananaProVertexNode is the Vertex AI edition of the same Gemini image call, and for most people the honest answer is: don't use it. The plain Nano Banana Pro node with an AI Studio key is dramatically less setup. This one exists for the people who are already living in Google Cloud - org-managed credentials, billing via the project, one billing account covering everything - and who either can't or don't want to hand out API keys.

The model is the same gemini-3-pro-image-preview (Nano Banana Pro, Google's 4K flagship), with the same 14 reference-image slots, same image_count up to 15, same 1K/2K/4K image_size options, same aspect_ratio list, and the same IMAGE batch output. The source comments flag one Vertex-specific quirk: this model wants the global region, which is why location defaults to it. Run it from a regional endpoint and you may get errors that make no sense until you remember that.

What's different

Three required inputs replace the single API key, and they're the whole story:

• project_id - your Google Cloud project ID, the one with Vertex AI enabled and billing on.
• location - defaults to global; the dropdown has ~40 regions but for this model you want global.
• service_account - a path to a service account JSON key file on disk. The node sets GOOGLE_APPLICATION_CREDENTIALS to it and builds a vertexai=True client.

Everything else - temperature, top_p, seed, system_instruction, the image_1–image_14 slots, use_concurrency, and the *_input override ports - is identical to the non-Vertex node. There are matching project_id_input, location_input, and service_account_input ports if you want an upstream node to supply them.

Installing it

Same pack, same install as the rest of ComfyUI's zhanglongxiao111/Comfyui-geminiapi:

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/zhanglongxiao111/Comfyui-geminiapi.git
cd Comfyui-geminiapi
python -m pip install -r requirements.txt

Restart and it's under ExternalAPI/Image/Generation next to its sibling. Note the pack's requirements.txt pulls the google-genai SDK, and this node needs it at 1.0.0+ or it refuses to load.

The gotchas

Setup friction aside, the failure behavior matches the AI Studio version: failed requests get padded out with black placeholder frames and a [Nano Banana Pro Vertex] Request failed: line in the log, so watch the console. The usual Vertex confessionals apply - the Vertex AI API must be enabled on the project, billing must be active, and the service account needs image-generation permissions. If a run "works" and returns only black frames, that's a failed request, not a moody model. And the same cost math holds: this is a paid API, SynthID-watermarked and censored at the source. Reach for it when your org already runs on GCP; otherwise the plain NanoBananaProNode gets you the same images with maybe twenty minutes less yak-shaving.