Concurrent Submit | zhenzhen-image-nb-lowprice

Nano Banana is the model name people actually google, and this node is where the pack serves it. zhenzhen-image-nb-lowprice bundles all four Nano Banana tiers into one node - flash, nb-2, nb-2-lite, and pro - and its Concurrent Submit wrapper queues them through the shared 30-slot image pool.

Nano Banana is Google's closed Gemini-image family: the consumer base product, the speed-quality hybrid, the lite tier, and the flagship Pro that does 4k native output and multi-image editing. None of it runs locally - no weights exist for you to download. The pack's whole pitch is reseller access to exactly this: the closed frontier, at "lowprice" rates. The README explicitly tracks the risk-control drama around these models (Google throttling, "优质分组" premium token groups being needed for stability), which is a preview of the troubleshooting section below.

How the wrapper works

Same as every node in the family: this Submit node mirrors the original's inputs, validates them, and hands the job to a bounded thread pool (30 image workers, tunable via COMFLY_IMAGE_CONCURRENCY). Output is a task (COMFLY_IMAGE_FUTURE) → Concurrent Collect Images (30) → IMAGE tensors in connection order + status.

The four models are all behind the same dropdown, which is convenient and slightly dangerous: switching models auto-swaps the available resolution/ratio lists, per the README's 20260726-1 note. Pro expects different shapes than flash, and the widget set changes to match.

Inputs that matter

• model - zhenzhen-image-nb-flash, zhenzhen-image-nb-2, zhenzhen-image-nb-2-lite, or zhenzhen-image-nb-pro. Flash and lite are the cheap/fast end; pro is the quality end that can do 4k.
• prompt - the text. Multiline.
• resolution - 0.5k, 1k, 2k, 4k. Yes, 0.5k is an option - that's the dirt-cheap draft tier. Resolution is your cost dial.
• size - fifteen options including auto and the extreme wide formats (1:4, 1:8, 4:1, 8:1). auto follows your reference image's ratio, which is what you want for edits.
• n - 1–4 images per call. Nano Banana's API caps at 4; the pool gives you parallelism across calls instead.
• image1–image14 - up to 14 reference images for multi-image editing. That's the Pro-class feature: paste a character across 14 refs and it holds.
• api_config - zhenzhen-seedance2-low-price-api-settings, domestic key from api.seedance.nz.
• seed - cache-control only, not a real sampler seed.

Install

ComfyUI Manager → search "Comfyui-zhenzhen" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

No downloads, no VRAM. API key required.

The Nano Banana–specific gotchas

• Google throttling is real, and it's not your fault. The README's long log is a blow-by-blow of Google cutting concurrency, 429/500 errors, and white-image failures on the default token group. The fix when the default group flakes: log into the site, edit your token, add the premium group ("优质分组"). The node can't fix server-side throttling - that's the price of a closed model through a reseller.
• 4k through the default group can return white images. The README's advice: prefer 2k on default, or switch to the premium group for 4k. A queued batch of 30 white images is a waste of a bill.
• The pool doesn't fix latency. Thirty parallel calls are still thirty calls to Google's rate-limited endpoint. When the site is under load, expect failures and re-runs, not magic.
• 14 refs × 30 slots = a lot of uploads. Reference uploads cost extra in this pack. Batch consciously.

This is the node to reach for when the job is "Google's best closed image model, in bulk." Just respect the resolution dial and keep the premium group on standby for the 4k work.