并发提交｜Zhenzhen Image Nano Banana 生成/编辑

Let's get the naming straight, because it matters: Nano Banana is Google's closed image model, and the zhenzhen-image-nb-* models in this pack are not it. They're the shop's own image models wearing the famous name, the way a reseller's catalog will reach for whatever word is trending. None of these four models - zhenzhen-image-nb-flash, zhenzhen-image-nb-2, zhenzhen-image-nb-2-lite, zhenzhen-image-nb-pro - runs Google's weights. If you came here expecting Gemini's 14-reference thinking mode, adjust expectations; what you get is a four-model image family that does text-to-image and reference-image editing, served per-call through api.seedance.nz.

That said, it's a well-behaved batch node. The model dropdown changes which knobs are live:

• zhenzhen-image-nb-flash - fixed 1k, single image per request, prompt capped at 1,000 characters. The fast tier.
• zhenzhen-image-nb-2 - 0.5k/1k/2k/4k, single image, and it unlocks the extended ratios (the 1:4/4:1/1:8/8:1 family).
• zhenzhen-image-nb-2-lite - fixed 1k, and the only model where n goes 1 to 4.
• zhenzhen-image-nb-pro - 1k/2k/4k, single image, standard ratios.

All four take up to 14 reference images (image1 … image14) for edit mode - connect any, and the node uploads them in slot order and submits them as editing references. n behaves per-model: only lite gets a range, the rest are fixed at 1. Output is the single future (SEEDANCE_IMAGE_FUTURE) into 并发接收图片（30 路）.

How it works

The family-standard concurrent pipeline: validate against the selected model, hand to the image thread pool (up to 30 workers; SEEDANCE_IMAGE_CONCURRENCY), return the future. References upload, the per-model payload assembles, the receiver collects in slot order. skip_error per slot, failure_mode on the receiver. The model-aware input limits (which resolutions/ratios/n are valid) are enforced before anything costs you a credit, which is worth more than it sounds when you're batching.

Installing it

ComfyUI Manager → search ComfyUI Seedance, or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/ComfyUI_Seedance.git
python -m pip install -r ComfyUI_Seedance/requirements.txt

Restart, look under Seedance/并发提交. requests and an api.seedance.nz key; nothing to download.

Where people get burned

• The name. It says Nano Banana, it isn't Nano Banana. You won't be burned by that except when you expect Gemini-grade prompt understanding and get a generic image model - prompt it like the mid-tier model it is.
• Per-model limits are real. Set resolution to 4k on nb-flash and the preflight stops you. The valid range depends on the dropdown; the tooltips spell each one out.
• Fourteen reference slots are a lot of upload. Same throttle-and-wait behavior as the pack's other big-slot nodes; trim before you wire.
• future → receiver, not Save Image. The family rule.

Per-image metering and the data-leaves-the-machine caveat apply. If you need actual Google Nano Banana, that's a different pack entirely - but if you want a batchable, mid-priced image model with a familiar name in the dropdown, this is it.