This Neon Function generates an image through Neon AI Gateway, stores the JPEG in Neon Object Storage, and returns a presigned URL.

Scaffold it

Install the latest Neon CLI, then create the function from the registry:

npm install -g neon@latest
neon login
neon link
neon functions new image-generation --slug imagegeneration

The CLI writes functions/imagegeneration/index.ts and declares the function in neon.ts.

Enable AI Gateway and declare the private images bucket:

neon config add ai-gateway
neon config add bucket images

You can also add both to neon.ts:

import { defineConfig } from "@neon/config/v1";

export default defineConfig({
  aiGateway: true,
  buckets: {
    images: { access: "private" },
  },
  functions: {
    imagegeneration: {
      name: "AI image generation with Neon Storage",
      source: "functions/imagegeneration/index.ts",
    },
  },
});

AI Gateway and Object Storage require a project in a supported region. AI Gateway also requires a paid Neon plan with prepaid credits. See AI Gateway get started and Object Storage get started.

Deploy once so Neon creates the bucket and injects the AWS_* and NEON_AI_GATEWAY_* variables into .env.local:

neon deploy

Run locally

neon dev

Call the URL that neon dev prints (the default is http://localhost:8787/):

curl -X POST http://localhost:8787/ \
  -H "content-type: application/json" \
  -d '{"prompt":"An astronaut drinking coffee in a neon-lit cafe, cinematic lighting"}'

Example response:

{
  "prompt": "An astronaut drinking coffee in a neon-lit cafe, cinematic lighting",
  "url": "https://<branch-id>.storage.<cell>.<region>.aws.neon.tech/images/generated/<uuid>.jpg?...",
  "key": "generated/<uuid>.jpg",
  "bucket": "images",
  "contentType": "image/jpeg",
  "bytes": 48213,
  "expiresIn": 3600,
  "model": "gpt-5-mini",
  "imageModel": "gpt-image-2"
}

The presigned URL is valid for one hour. The bucket is private, so the object has no permanent public URL. Reuse the object key with the S3 API to sign a new URL when you need to share it again.

Request format

FieldRule
promptRequired. A non-empty string up to 2,000 characters.

The function only accepts POST /.

This template doesn't authenticate requests, so anyone with the function URL can call it and spend AI Gateway credits. Add authentication before you expose it.

The template uses gpt-5-mini to call the gpt-image-2 image-generation tool, which makes a 1024x1024 low-quality JPEG. Change MODEL, IMAGE_MODEL, or the tool options in functions/index.ts to tune cost and output.