Zhenzhen Image Nano Banana 生成/编辑

Nano Banana is Google's Gemini-native image model - the one with the banana emoji as its entire marketing strategy, and one of the most in-demand closed models of 2026. Zhenzhen_Image_NB is that model's doorway into T8mars/ComfyUI_Seedance: four Nano Banana variants under one dropdown, text-to-image and reference-image editing with up to 14 reference images, all served through the pack's reseller API rather than Google's own.

The "reseller" part deserves a beat, because it changes your expectations. Google's Nano Banana is heavily censored - the KB notes the ImageFX/API route has strict IP, celebrity, and safety filtering - and the model's refusals happen in the model, not the node. A reseller can't bypass that; it just gives you another billing/region door in. If you're in the pack for video and want Google-grade image quality on the same bill, this node is the fastest way there. If you're shopping for Nano Banana specifically, weigh the official path (prepaid Comfy credits or the Gemini API) against this - you're trading a vetted route for a single-key reseller one.

How it works

Standard pack mechanics: upload, submit to /v1/image/generations, poll, download, decode. The model dropdown picks the variant, and each one has different limits - the node's dynamic UI enforces them:

• zhenzhen-image-nb-flash (default) - fixed 1k, fixed count of 1, prompt capped at 1000 characters.
• zhenzhen-image-nb-2 - resolution 0.5k/1k/2k/4k, fixed count of 1, extended aspect ratios up to 1:4/4:1/1:8/8:1.
• zhenzhen-image-nb-2-lite - fixed 1k, but n runs 1 to 4 (the only model here that returns multiple images).
• zhenzhen-image-nb-pro - 1k/2k/4k, fixed count of 1, standard ratios.

All four support text-to-image with no references, or editing with image1–image14 connected. No reference image = generation; references connected = editing. Note the pack downloads the primary gateway result even on models that support n.

Inputs and outputs that matter

• model - the four-variant switch above.
• prompt - required; 1000-character cap on flash.
• resolution - model-specific subset of 0.5k–4k.
• size - model-specific ratios, including extended 1:4–8:1 formats on nb-2/lite.
• n - 1 to 4, but only nb-2-lite honors it; others fix at 1.
• image1–image14 - optional reference images for editing.

Outputs: image, image_url, task_id, response.

Install

One node in T8mars/ComfyUI_Seedance:

# ComfyUI Manager: search "ComfyUI Seedance", or:
comfy node install seedance

Or clone https://github.com/T8mars/ComfyUI_Seedance.git into ComfyUI/custom_nodes, pip install -r requirements.txt, restart, and configure a key from api.seedance.nz/console via the Seedance API Config node or SEEDANCE_API_KEY.

Where people get burned

The model-specific limits are the trap - set n=4 on nb-flash and it won't give you four images, and a 2000-character prompt on flash gets rejected at the 1000 cap. Also remember these are Google's models under the hood: the famous Nano Banana censorship applies no matter how "loose" the reseller's marketing claims to be, and your prompts and reference images go to a third party's server. If you want the extended aspect ratios, you have to be on nb-2 or lite. Start with flash at 1k - it's the cheapest smoke test - and only escalate to pro when the flash result isn't good enough.