JILIGULU Nano Banana Fast

If you want Google's Nano Banana (the Gemini-native image family) inside ComfyUI without paying for a 4K flagship, this is the speed-first option - and it's the one with the trap built in. JILIGULU Nano Banana Fast is an API-client node for nano-banana-fast on jiligulu.art, a cloud aggregator that resells the Nano Banana lineup. The whole point, per the service's own description, is fast rough drafts, quick multi-option exploration, and low-cost batch content production. No local GPU, no model files, just a prompt (and at most one reference image) sent to a remote API.

That single-reference cap is the thing that catches people. The other nodes in this pack accept 8–14 reference images; this one accepts exactly one, and the node enforces it locally before the request ever leaves your machine. Load a batch of two images and it fails on the spot.

How it actually works

The mechanism is identical to its siblings in the JILIGULU pack - same base class under the hood. It fetches the model config, uploads your (single) reference as a PNG, posts a run job, polls the task endpoint every 5 seconds for up to 15 minutes, then downloads the result and hands you a ComfyUI IMAGE. Zero new Python dependencies: it runs on the standard library plus numpy/torch/PIL, which ComfyUI already ships.

The inputs that matter

• prompt - required, as usual; an empty prompt is a hard error.
• api_key - paste it in the node, or set JILIGULU_API_KEY before starting ComfyUI.
• aspect_ratio - auto plus the standard 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 5:4, 4:5, and 21:9.
• image_size - 1K, 2K, or 4K.
• image - the optional reference input, and the one that matters here: one image only. Anything beyond a single frame and the node refuses to run.

Output: image (IMAGE). Feed it to a preview or save node, or into an upscaler if you're drafting at 1K and want to see what the composition becomes at print size.

Installing it

ComfyUI Manager → search "JILIGULU Nano Banana" → install → restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/XYQC-LH/ComfyUI-JILIGULU-Nano-Banana

Restart ComfyUI and you're done - no requirements.txt to resolve, no model to download.

Common issues

• "Too many reference images." This is the node's signature failure. You feed it two refs, it errors locally. That's the Fast tier doing its job - keep it to one image and use the base, Pro, or 2 nodes when you need more.
• Chinese error messages. The author raises errors in Chinese. They read accurately once you translate them; the most common is a missing API key ("缺少 api_key").
• Slowness isn't the node's fault. It polls up to 15 minutes per job. Fast means the model tier is speed-oriented, not that ComfyUI returns instantly.
• It's a paid, obscure service. Every run spends credits with a Chinese aggregator that has essentially zero presence in the English community, and this pack is brand new. The underlying model is Google's, which means aggressive content filtering and an invisible SynthID watermark. Test cheap, read the terms, don't route sensitive material through it.

Use it for what it's for - rapid idea roulette at low cost - and it's a fine addition to the graph. Expect multi-reference character work out of it and you'll just be annoyed.