Introduction

The Sprite AI API turns text prompts into game-ready sprites and sprite-sheet animations. Use it from a script, a build step, an MCP server, or an editor plugin.

Requests are JSON, authenticated with a Bearer key. Generated images come back inline as base64, so there's no file to download separately.

What you can do

Generate or import once and you get a sprite id. Animating a sprite (source_generation_id) and style-matching a new one to it (reference_asset_id) both take that id, so you never resend image bytes. To use a local file, import it first.

1. Create a key

Open the API keys tab and generate one. It starts with sai_sk_ and is shown once, so copy it now.

2. Send a request

POST a prompt and an asset type with your key as a Bearer token.

curl https://www.sprite-ai.art/api/sprites \
-H "Authorization: Bearer sai_sk_your_key_here" \
-H "Content-Type: application/json" \
-d '{ "prompt": "a knight with a blue cape", "asset_type": "character" }'

const res = await fetch("https://www.sprite-ai.art/api/sprites", {
method: "POST",
headers: {
  Authorization: "Bearer sai_sk_your_key_here",
  "Content-Type": "application/json",
},
body: JSON.stringify({
  prompt: "a knight with a blue cape",
  asset_type: "character",
}),
});
const data = await res.json();

import base64, requests

res = requests.post(
  "https://www.sprite-ai.art/api/sprites",
  headers={"Authorization": "Bearer sai_sk_your_key_here"},
  json={"prompt": "a knight with a blue cape", "asset_type": "character"},
)
data = res.json()

3. Save the image

The PNG is in image.pngBase64. Decode it and write it to a file.

echo "iVBORw0KGgoAAAANS..." | base64 -d > knight.png

import { writeFile } from "fs/promises";

await writeFile("knight.png", Buffer.from(data.image.pngBase64, "base64"));

with open("knight.png", "wb") as f:
  f.write(base64.b64decode(data["image"]["pngBase64"]))

Sync vs async

Still sprites are synchronous: one request, the finished image in the response. Animations take about 90 seconds, so they run as a job you start and then poll until it's done.

You don't pick a model. Describe what you want, and the request is routed for you based on the asset type and size.