Nano Banana 2 Lite

What it is

A stripped Nano Banana 2. Three widgets, one of them optional, no resolution tier, no output format, no seed. Prompt, ratio, generate.

That sounds like a downgrade until you look at what it's for. Nano Banana models are closed - no weights, no local option, per-call billing forever (docs/knowledge/closed-source-models.md). Every iteration costs money. So a node with fewer knobs and a lower price tier is genuinely useful as the sketch stage of a pipeline: rough in the composition cheap, then re-run the keeper through Nano Banana 2 or Nano Banana Pro with ratio, resolution and format actually chosen. Nobody paints the underdrawing on canvas they're going to sell.

How it works

The node is generated from KIE's request schema for the Lite endpoint, which is why it's so sparse - the pack mirrors KIE's documented fields rather than inventing a UI. Your prompt (and optionally an image) are sent as a generation task, the node polls it until completion or timeout_seconds (1200 by default), and the result comes back as a native ComfyUI IMAGE. Prompts of up to 20,000 characters are accepted, so length isn't the constraint here.

image_urls is a real IMAGE socket, so the pack uploads whatever you connect to KIE's storage and substitutes the URL. No hosting, no pasting links.

Inputs and outputs

Required:

• prompt (STRING) - ships pre-filled with "Generate a pig on the grass, cinematic light", which is a perfectly good demonstration and a terrible thing to leave in place. Clear it.
• aspect_ratio - auto by default, plus fifteen options including the very tall and very wide ones (1:8, 8:1, 4:5, 5:4). auto means the system picks; pick explicitly if the output feeds a fixed-size layout.

Optional:

• image_urls (IMAGE) - reference/edit input. A batch is fine.
• timeout_seconds (default 1200, i.e. 20 minutes) and callback_url - leave the callback empty for normal ComfyUI use; it's for server-side webhook setups.

Outputs: image, url, all_urls_json, task_id, raw_json, credits_consumed, credits_left. Same socket set as the rest of the pack's image nodes, so you can swap Lite for Nano Banana 2 in a graph without rewiring anything.

Install

Manager: search KIE-Nodes-Next, install, restart ComfyUI. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next

Then Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key. The key is validated before saving; you get a masked fingerprint and, when KIE's balance service is up, your remaining credits - handy for spotting that you're out of credit rather than broken. KIE_API_KEY in the environment takes priority over the saved key for server installs.

Dependencies are requests, Pillow, numpy, PyYAML. There's also a one-click Windows .vbs installer in dist/ that finds Desktop/Portable/manual installs; use a v0.3.1-or-later build, since the earliest version could leave backup folders inside custom_nodes for ComfyUI to import as nodes.

Where people get burned

Expecting the flagship. "Lite" is doing real work in that name. For text rendering and prompt-heavy compositions you want Pro; this one is for shape, mood and composition checks.

No resolution or format control. Whatever the endpoint defaults to is what you get. If your downstream pipeline assumes PNG with alpha, run the final through a node that lets you choose.

auto aspect ratio surprises. Same story as the rest of the family: fine for a mood board, annoying when a subsequent node expects a size.

The prefilled prompt. It's a demo, and it will happily generate you a picture of a pig on grass at your expense.

Retired endpoints. KIE's catalog is live and changing - a node can start failing because the provider moved or dropped an endpoint, not because your workflow changed. KIE • Refresh Model Catalog re-syncs the pack against KIE's current docs, and there's an offline bootstrap in the meantime.

That this is a reseller. KIE proxies Google's model; the model's own content filters still apply regardless of what the node exposes. No node-side setting changes that.