Nano Banana Pro (FAL)

Nano Banana Pro (FAL) is a ComfyUI wrapper for Google's Gemini 3 Pro Image - the model everyone knows as "Nano Banana Pro," running on fal.ai's infrastructure. The KB's closed-source coverage sums up why this node exists: Nano Banana Pro is 4K-native, can take up to 14 reference images, and does "thinking mode" reasoning before generating. It's also entirely closed and API-gated. There's no weights file to download, ever. This node hands you a Google model through a FAL account instead of a Gemini billing page.

The mechanism is worth knowing because it's the one node in the pack that rolls its own queue handling. Rather than the standard fal client call, the source submits directly to queue.fal.run/fal-ai/nano-banana-pro/edit, grabs a request_id, and polls the status endpoint itself - with a 75-second timeout. When it times out, it doesn't error: it writes a black placeholder image and records the timeout in generation_info. That's a design choice you'll want to know about, because "I got a black image back" is not the failure it looks like - it's the node's way of saying the API took too long.

The inputs, from the schema:

• api_key - FAL key (node input or FAL_KEY env var).
• prompt (default "A beautiful masterpiece") - with Nano Banana Pro, be descriptive; this is the model the KB describes as powerful but "unimaginative" without a real prompt behind it.
• num_images (1–4) - how many outputs per run.
• image1 through image4 (image1 required) - the reference images. Up to four here, versus the 14 the model can theoretically eat; this node caps it.
• aspect_ratio - custom, 16:9, 4:3, 1:1, 3:4, 9:16, 21:9.
• resolution - 1K, 2K, or 4K. The 4K tier is the flagship capability and the most expensive.
• seed - optional.

Outputs: images (a batch, all generated results), image_paths (semicolon-joined paths), generation_info, generation_time.

Why you'd reach for it: Nano Banana Pro is one of the top rungs of the KB's upscaling ladder - Comfy Org's own upscaling handbook names it as one of the three image upscalers that matter, next to Magnific and SeedVR2 - and its multi-image reference editing is the model's party trick: give it a few images of a subject and it composites them into something new. This node is the way to run that without a Gemini subscription. It's also the pack's most "editing-heavy" generation node: this is image generation from references, not just text-to-image.

The honest caveats. Cost: 4K multi-image runs are the priciest things in this pack, and the 75-second internal timeout means long 4K jobs can hit the black-image fallback before the model finishes - if you see black, retry and consider dropping to 2K. The KB's "unimaginative" label is the other one: Nano Banana Pro is a stunning tool and a mediocre idea-generator, so bring the concept yourself and let it do the rendering.

Install is the pack standard: Manager search "Node Fun", or clone https://github.com/nofunstudio/Node_Fun_ComfyUI into custom_nodes, pip install -r requirements.txt, restart, add a FAL key. Outputs save to output/API/FAL/ with metadata beside them. If you've been eyeing Nano Banana Pro from the closed-source sidelines, this is the least-friction seat.