PVL Google NanoBanana PRO Multi API

This is the top of the Nano Banana food chain. Where the rest of this pack's Google nodes run Gemini 2.5 Flash Image, this one runs Gemini 3 Pro Image - the model Google markets as Nano Banana Pro: 4K native output, up to 14 reference images (this node exposes eight), and "thinking" before generation. It's the flagship, it's the priciest, and if you're chasing the best closed-model output available through an API, this is where you go.

The pack's signature resilience carries over: use_fal_fallback defaults on, routing to fal's nano-banana-pro / nano-banana-pro/edit routes when Google stumbles, with force_fal to skip Google entirely. There's also use_google_search - a toggle that grounds the image generation with live web search results, which is a genuinely unusual feature and a sign of how far the API has come. It's useful for "generate an image of [current event] in this style" type prompts, at the cost of latency and, presumably, your patience.

How it works

Same architecture as the rest of the family: connected reference images (up to eight) plus your prompt go to the Gemini 3 Pro Image model; [++]-delimited prompts split into parallel generations; resolution (default 1K) selects output scale; fal fallback catches failures. The text output captures the model's response when capture_text_output is on.

The inputs that matter

• prompt - the generation brief.
• model - default gemini-3-pro-image-preview.
• resolution - 1K by default; 2K/4K are where this model's edge lives (4K is the flagship spec), and they cost more.
• image_1 … image_8 - reference images for multi-image control.
• num_images - up to 16 per run, the most of any node in this pack.
• use_google_search - off by default; flip on for fact-grounded or current-scene generation.
• aspect_ratio - default 1:1.
• use_fal_fallback, force_fal, fal_api_key, fal_route_txt2img (fal-ai/nano-banana-pro), fal_route_img2img (fal-ai/nano-banana-pro/edit).
• api_key, capture_text_output, seed, output_format, retries, timeout_sec, debug_log, delimiter, sync_mode, endpoint_override, request_id.

Outputs: text (STRING) and images (IMAGE).

Installing it

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/pvlprk/comfyui-pvl-api-nodes

Restart, export GEMINI_API_KEY and FAL_KEY, or fill the fields. No downloads.

Common issues

Let's be straight about cost: this is the most expensive node in the pack to run, and the KB's pricing picture puts Nano Banana Pro anywhere from $0.039 to $0.24 per image depending on resolution. At 4K with num_images at 16, one execution can be a real line item. Treat it as a quality ceiling you visit deliberately, not a default. Second, use_google_search is off by default for a reason - it adds latency and its output is only as good as the search results; leave it off until you have a prompt that clearly needs it. Third, the thinking-mode behavior means responses are slower than Flash Image even when it works, so don't diagnose "the node is stuck" - wait. And the filtering is the same family policy: celebrity/IP and sensitive-content prompts get blocked, which is exactly when the fal fallback engages (and bills you).