One brief, four image models, four prices

Open it in Replit with one click, add your own key, and change the brief.

Type a brief once. It goes to grok, nano-banana, gpt-image-2 and nano-banana-pro at the same time, and you get four takes back with what each one actually charged. Around 48 cents for the set.

model	price per image
grok	$0.07
nano-banana	$0.07
nano-banana-pro	$0.13
gpt-image-2	$0.21

Quoted 2026-09-08 at 1:1 and 1K. Prices are set by the provider and can change, and they depend on the arguments you send, so a bigger imageSize costs more. The app shows you what each call actually cost, which is the only number that is ever authoritative.

The point is not a cheaper subscription. It is a wider bench. Nobody can tell you in advance which model reads your brief the way you meant it, so ask four and pick the one you like.

Run it

1. Open it in Replit. That imports this repo into your own account as a runnable copy.
2. Get a key at app.taskfuel.ai. The first $1 is on the house, which is about 2 runs of this app.
3. Open the Secrets tab in the left sidebar. Add a secret named TASKFUEL_API_KEY and paste your key as the value.
4. Hit Run.

That is the whole setup. No provider accounts, no per-model API keys, no subscription to cancel.

How it works

Every paid call goes to one endpoint:

await fetch("https://app.taskfuel.ai/v1/call", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.TASKFUEL_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    url: "https://stablestudio.dev/api/generate/nano-banana/generate",
    method: "POST",
    body: { prompt, aspectRatio: "1:1", imageSize: "1K" },
    maxAmountUsd: 0.25,
  }),
});

TaskFuel pays the provider's HTTP-402 charge from your prepaid balance and passes the response straight back. The response headers tell you what it cost (x-taskfuel-cost) and what is left (x-taskfuel-balance).

Generation is asynchronous: the first call returns a jobId, and server.js polls a free status endpoint until the image is ready.

Spending safely

The key can spend your whole balance, and nothing is watching at call time, so the limits live in the code:

Setting	Default	What it does
MAX_USD_PER_IMAGE	0.25	Hard ceiling on any single image. The gateway rejects anything above it.
DAILY_BUDGET_USD	2.50	Stops generating once the day's spend hits this.

Both are optional secrets you can change without touching the code.

If you make this public and let strangers use it, they are spending your balance. Keep the budget low, or make each visitor bring their own key.

Make it yours

• Swap the models. Edit MODELS in server.js. The full list is at stablestudio.dev, and there are video models and an image-to-SVG endpoint in there too.
• Change the shape. aspectRatio takes 16:9, 9:16, 4:3 and more. imageSize goes up to 4K.
• Find something else entirely. GET https://app.taskfuel.ai/v1/discover?q=... searches every provider in the catalog. There are around 90 of them, covering search, market data, email, phone calls and more.

Full guide for wiring an app to the gateway: app.taskfuel.ai/building-apps.md

Running outside Replit

cp .env.example .env   # then put your real key in it
node --env-file=.env server.js

Needs Node 20 or newer. There are no dependencies to install.