Documentation

API reference

AI placeholder images from a URL. Drop a link into an <img> tag and get a contextually appropriate, web-ready image — generated once, cached forever on a CDN.

Base URL for this instance: https://your-instance.vercel.app

Quick start

Paste this into any HTML page and an image appears — no API key, no JavaScript:

<img src="https://your-instance.vercel.app/i/1200x600?prompt=modern+fintech+dashboard+hero&style=photographic" />

That's the entire integration. The first load generates the image; every load after is an instant CDN cache hit.

The URL pattern

https://your-instance.vercel.app/i/{width}x{height}?prompt={description}&style={style}

{width}x{height} — exact pixel dimensions, e.g. 1200x600
prompt — what the image should show (URL-encode spaces as +)
style — optional visual preset (see below)

Parameters

Param	Required	Default	Notes
`prompt`	required	`—`	Image description, max 1000 chars
`style`	no	`web`	Visual preset — see Styles
`seed`	no	`0`	Change for a different image, same prompt
`fmt`	no	`webp`	webp \| jpeg \| png
`q`	no	`82`	Quality 1–100 (ignored for png)

Dimensions are clamped to 64–2048 per side, ~4 megapixels max.

Styles

A server-side prompt layer turns terse prompts into clean web imagery. Select with style=:

Style	Description
`web`	Default — neutral, clean, professional. Reads well behind text.
`photographic`	Realistic photo style, natural lighting.
`illustration`	Flat / semi-flat digital illustration.
`abstract`	Flowing shapes, decorative backgrounds.
`3d`	Photoreal 3D render, studio lighting.
`minimal`	Ample white space, muted palette.

Examples

<!-- Hero banner -->
<img src="https://your-instance.vercel.app/i/1600x700?prompt=modern+SaaS+product+hero+dashboard&style=photographic" />

<!-- Team / about photo -->
<img src="https://your-instance.vercel.app/i/800x600?prompt=diverse+team+collaborating+in+bright+office&style=photographic" />

<!-- CSS background -->
.hero {
  background-image: url("https://your-instance.vercel.app/i/1920x1080?prompt=abstract+blue+gradient+tech+background&style=abstract");
  background-size: cover;
}

<!-- Different image, same prompt -->
https://your-instance.vercel.app/i/600x400?prompt=coffee+shop+interior&seed=2

JSON API

When you want the URL handed back instead of generating on load:

curl -X POST https://your-instance.vercel.app/api/generate \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "modern fintech dashboard hero",
    "width": 1200,
    "height": 600,
    "style": "photographic"
  }'

Response:

{
  "url": "https://...blob.vercel-storage.com/cache/abc123.webp",
  "cached": false,
  "width": 1200,
  "height": 600,
  "model": "gemini-2.5-flash-image",
  "id": "abc123..."
}

Health

Liveness check and the active model:

curl https://your-instance.vercel.app/api/health
# → { "ok": true, "model": "gemini-2.5-flash-image" }

Using with Claude Code

Drop this into your project's CLAUDE.md and Claude Code uses Banana Stand automatically whenever it adds images:

# Placeholder Images

For placeholder images on this site, use the Banana Stand API.
Write image URLs as:

https://your-instance.vercel.app/i/{width}x{height}?prompt={url-encoded+description}&style=photographic

Place them directly in <img src> or CSS background-image. Choose
dimensions that match the layout slot. Same prompt + size always
returns the same cached image, so reuse URLs for stable pages.

Styles: web (default), photographic, illustration, abstract, 3d, minimal.
Tips: encode spaces as +, add &seed=2 for a variant.

No SDK, no keys in the page. Claude writes the URL, the page renders, the service does the rest.

Access control

Generating a new image calls Gemini and costs money. Set a GEN_TOKEN env var on your instance to gate new generations:

Cache hits are always open — already-generated URLs serve to anyone, no token. Embedded <img> tags never break.
Cache misses require a matching X-Gen-Token header when GEN_TOKEN is set, otherwise they return 401.

# Generating a new image only works with the secret
curl -H "X-Gen-Token: YOUR_SECRET" \
  "https://your-instance.vercel.app/i/1200x600?prompt=team+in+a+bright+office"

Set GEN_TOKEN in your Vercel project (Production scope) and redeploy. Leave it unset for a private/self-host instance where anyone may generate.

Caching & limits

Deterministic: same params → same image, forever. Pages stay stable across reloads and deploys.
Cache hitsare instant CDN redirects (~150ms) and don't call Gemini.
Cache misses generate in ~2–10s, then are cached.
Generation control: set GEN_TOKEN to require X-Gen-Token on new generations (see Access control). Cache hits are always open.

⚠️ Images are AI-generated and carry an invisible SynthID watermark. Don't present them as authentic photography.

Try the test page →View on GitHub →