🍌 智绘香蕉2 (极速版)
Nano Banana 2 in your graph via a reseller API — key required, no local GPU
- image_1
- image_2
- image_3
- image_4
- 生成图片
- 详细信息
- 智绘显示
ZH_BananaTutuPort (🍌 智绘香蕉2 极速版) is an API-wrapper node for Google's Nano Banana 2 - the closed Gemini image model - plus a Gemini 3 Pro preview option, and it gets you there by routing through a third-party endpoint. If the phrase "closed-source model" makes you reach for the wallet, that's the whole pitch: you can't download Nano Banana 2, so you call it. This node is the call, packaged so the returned image drops into your ComfyUI graph exactly like a local generator would. The name "TutuPort" is a port of a "Tutu"-style banana API wrapper; the "(极速版)" is about the session reuse under the hood.
What it actually is
Read this closely before you wire it up: the node does not run anything locally. It's a JSON HTTP client with your API key baked in. It POSTs your prompt, aspect ratio, size and seed to base_url - default https://ai.t8star.cn/v1/images/generations, a reseller/proxy endpoint, not Google's first-party API - and decodes the base64 image that comes back. t8star.cn is the kind of reseller the KB's API-nodes doc describes: cheaper than first-party, sometimes the only route in certain regions, and completely outside Google's billing and moderation pipe. Your key, your call, their endpoint.
It supports image editing too: optional image_1 through image_4 IMAGE sockets get base64-encoded and sent along as references, which is how you do nano-banana-style subject editing or multi-image conditioning without leaving the graph.
The inputs that matter
api_key- the only thing that makes it run. The placeholder sayssk-xxx..., and the node literally refuses with an error image if it's empty. Get it from whoever runs your endpoint.model-nano-banana-2(default) orgemini-3-pro-image-preview.prompt- what you want, as English as you'd write to Gemini.aspect_ratio(ten choices, 1:1 default),image_size(1K/2K/4K, 2K default),seed(0 = random), andimage_quality(JPEG compression, 60–100, default 85 - this one affects file size and compression artifacts, not model output quality).
Outputs are three: 生成图片 (IMAGE) to feed your save or upscale, 详细信息 (STRING) with the raw response detail, and 智绘显示 (STRING) - a pretty-printed status panel the pack renders for you. This is an output node, so it's built to be the end of a branch.
Install
Part of the 智绘灵箱 (ComfyUI-ZhiHui) pack:
cd ComfyUI/custom_nodes
git clone https://github.com/zhuyungen/ComfyUI-ZhiHui.git
Restart ComfyUI (or ComfyUI Manager, "智绘灵箱" / "ComfyUI-ZhiHui"). Only requests is needed beyond the basics.
Where people get burned
- No key, no picture - and it's silent-ish about it. Missing key produces an error image and console message, so check the console if you get a red 512×512 square.
base_urlis a live dependency. The default points at a third-party reseller whose endpoint can change, rate-limit, or vanish. If generation starts failing, the first thing to check is whether your endpoint is still up, not your prompt.- This is the security-shaped node. The KB's API-nodes doc puts it bluntly: a node whose whole job is to phone home with your API key is exactly the shape of past credential-stealing malware. That doesn't make this one malicious - but use a key scoped to one provider, and don't paste a high-value key into a shared workflow file.
- Image quality ≠ resolution. Crank
image_qualityto 100 and you've just inflated file size, not model detail. 85 is fine.
If you want Nano Banana 2 inside ComfyUI and you're already in this pack, this is the fastest door - but the KB's honest take applies: the official ComfyUI Partner Nodes route the same model through Comfy's own metered storefront, and which is safer is your call. This one is a key, an endpoint, and a prompt away from working.
Inputs (12)
| Name | Type | Default | Description |
|---|---|---|---|
| base_url | STRING | https://ai.t8star.cn/v1/images/generations | — |
| api_key | STRING | — | |
| model | COMBO | nano-banana-2 | 2 options: nano-banana-2, gemini-3-pro-image-preview |
| prompt | STRING | A cute futuristic cat | — |
| aspect_ratio | COMBO | 1:1 | 10 options: 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, +4 |
| image_size | COMBO | 2K | 3 options: 1K, 2K, 4K |
| seed | INT | 00–18446744073709550000 | — |
| image_quality | INT | 8560–100 | JPEG压缩质量 (60-100) |
| image_1opt | IMAGE | — | |
| image_2opt | IMAGE | — | |
| image_3opt | IMAGE | — | |
| image_4opt | IMAGE | — |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| 生成图片 | IMAGE | — |
| 详细信息 | STRING | — |
| 智绘显示 | STRING | — |