@blackbelt-technology/pi-dashboard-nano-banana

TypeScript port of the standalone nano-banana-imagegen pi skill — no Python. Generate and edit images with Google's Gemini image models via the @the-focus-ai/nano-banana CLI, wrapped with automatic GEMINI_API_KEY resolution, output-path handling and bounded-concurrency batch generation.

Exposed two ways:

• pi skill — .pi/skills/nano-banana-imagegen (auto-loads on image-gen requests).
• CLI bin — pi-nano-banana: a plain-JS launcher (bin/pi-nano-banana) that runs the TypeScript source via tsx in a checkout and an esbuild bundle (dist/nano-banana.js, built on prepack) from an npm install, so it starts from any shell. Library consumers keep importing src/*.ts through the exports map.

Usage

pi-nano-banana "a serene mountain landscape at sunset"
pi-nano-banana "add a hot air balloon to the sky" --file photo.jpg
pi-nano-banana "a minimalist logo" --output logo.png --model gemini-2.0-flash-exp

GEMINI_API_KEY (or GOOGLE_API_KEY) resolves from, in order: --api-key, the environment, a project-local .env (cwd + up to two parents), then a package-local .env. Nothing is committed — .env is gitignored.

pi backend (opt-in)

--backend pi (or NANO_BANANA_BACKEND=pi; programmatic backend: "pi") generates through pi's model runtime with the OpenRouter credential pi already holds — /login openrouter in pi, or OPENROUTER_API_KEY. No GEMINI_API_KEY needed.

• Opt-in only: the default stays gemini, a missing Gemini key never falls back to pi, and the library never reads NANO_BANANA_BACKEND (only the CLI does).
• Model ids are OpenRouter image ids: bare Gemini ids get google/ in front; non-Google models need the full vendor/model id. Default google/gemini-2.5-flash-image; --flash → google/gemini-3.1-flash-lite-image.
• The CLI prints (pi · <model>) · ~$<cost> est. — cost is an estimate from token usage.
• Needs @earendil-works/pi-coding-agent >= 1.0.0 resolvable next to the package (optional peer).

pi-nano-banana "a red fox in the snow, watercolor" -o fox.png --backend pi

Programmatic API

import { generateImage, batchGenerate } from "@blackbelt-technology/pi-dashboard-nano-banana/nano-banana.js";

await generateImage({ prompt: "a fox", output: "fox.png" });
await generateImage({ prompt: "a fox", output: "fox.png", backend: "pi" }); // → { ok, output, outputs, model, usage, backend }

await batchGenerate({
  jobs: [
    { name: "hero", prompt: "wide cinematic hero", output: "out/hero.png" },
    { name: "icon", prompt: "flat minimal icon", output: "out/icon.png" },
  ],
  concurrency: 3,
});

batchGenerate powers the storyboard step of @blackbelt-technology/pi-dashboard-video-production.

Prerequisites

• Node (runs as TypeScript via pi's jiti loader — no build step).
• Network access for npx to fetch the underlying nano-banana CLI.
• A Gemini API key from https://aistudio.google.com/apikey — or, for --backend pi, an OpenRouter credential in pi and @earendil-works/pi-coding-agent >= 1.0.0.