Run the same prompt through all three Google Flow Nano Banana image models — Nano Banana 2 Lite (nano-banana-2-lite, Gemini 3.1 Flash-Lite Image, the default), Nano Banana 2.1 (nano-banana-2), and Nano Banana Pro (nano-banana-pro, Gemini 3 Pro Image) — through the Google Flow API by useapi.net, and download every result labeled by model.
📖 Full walkthrough: Nano Banana 2 Lite vs 2.1 vs Pro: Google Flow Image Models Compared — July 2, 2026 (October 5, 2026)
google-flow-nano-banana-compare.mjs (Node.js) and google-flow-nano-banana-compare.py (Python) are equivalent implementations — each reads prompts.json (one entry per model) and calls POST /images. Because that endpoint is synchronous, the script writes each result image as soon as the call returns.
- Node.js v21 or newer (no dependencies — uses built-in
fetch), or Python 3.x (standard library only — no dependencies) - A useapi.net API token
- A connected Google Flow account email (image generation works with free Google AI accounts)
node ./google-flow-nano-banana-compare.mjs <API_TOKEN> <EMAIL> [PROMPTS_FILE]
python3 ./google-flow-nano-banana-compare.py <API_TOKEN> <EMAIL> [PROMPTS_FILE]PROMPTS_FILE defaults to prompts.json. The script looks the account up by email and checks its health field before submitting, and sends no captchaToken — the useapi.net worker solves the reCAPTCHA automatically. Output images are named <model>_<n>.jpg.
prompts.json is an array of prompt objects — one per model, all with the same prompt so you can compare interpretations. prompt is the only required field; model, aspectRatio, count, and seed fall back to the API defaults (model nano-banana-2-lite, four variations, 16:9). All three models accept up to 10 reference images (reference_1…reference_10), the auto aspect ratio in image-to-image mode, and 2K/4K upscaling.
Every parameter is documented on POST /images.