GPT Image 2 API Guide

This guide is about the API workflow inside the gpt-image2 project.

It is not pretending to be a separate official API portal. The goal is to show how the project handles image generation requests, what the request shape looks like, and when you should use the browser tool instead of writing integration code immediately.

If you only want to test prompt quality and output style, open the GPT Image 2 generator first. That is still the fastest route to value.

Two useful levels of integration in this project

There are two practical ways to think about the API layer here:

1. Quick generation route

Use this when you want to understand the basic request shape and test image generation quickly.

Route:

/api/generate-images

2. Full app job route

Use this when you need the application workflow with auth, credits, persisted jobs, and asynchronous provider execution.

Route:

/api/aiimage/start

Minimal request example

The simplest route in this project accepts a prompt, provider, model ID, and aspect ratio.

curl -X POST https://www.gpt-image2.dev/api/generate-images \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Editorial product shot of a silver camera on a soft gray desk, crisp studio lighting, premium ad style",
    "provider": "openai",
    "modelId": "dall-e-3",
    "aspectRatio": "1:1"
  }'

Example response:

{
  "provider": "openai",
  "image": "<base64-image-data>"
}

JavaScript example

const response = await fetch('/api/generate-images', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    prompt: 'Minimalist poster of a moonlit city skyline, clean contrast, cinematic mood',
    provider: 'openai',
    modelId: 'dall-e-3',
    aspectRatio: '3:2',
  }),
});

const data = await response.json();

When to use the authenticated route

The deeper route is useful when you want app-level behavior instead of a quick test.

Example payload shape:

{
  "featureId": "ai-baby-image",
  "mode": "text_to_image",
  "modelId": "gpt-image-2-t2i",
  "prompt": "Premium product render of a ceramic lamp on a walnut side table, warm interior lighting",
  "aspectRatio": "1:1"
}

This route is the better choice when you care about:

• user session validation
• credit usage
• stored jobs
• provider task orchestration
• later polling or retrieval

Which fields matter most

prompt

This is still the most important field. Be specific about subject, setting, and the visual direction you want.

provider and modelId

The project separates provider choice from model ID so the integration layer can change models without redesigning the whole request contract.

aspectRatio

Aspect ratio affects both framing and output size behavior, so it is better to pass it explicitly instead of letting the request guess.

When not to use the API first

A lot of people search for gpt image 2 api when they are still figuring out prompts and output quality. In that situation, code is usually not the bottleneck.

If you are still doing any of the following:

• testing styles
• comparing prompt patterns
• deciding whether the output quality is good enough
• learning how reference edits should work

then the GPT Image 2 generator is still the better first stop.

If you searched for image 2.0 api, chatgpt image2 api, or Imagen 2 API

These terms show up together in search, but they do not all mean the same thing.

image 2.0 api

This is usually shorthand for a newer GPT-style image workflow. The right response is to explain the actual request format the project supports.

chatgpt image2 api

This usually means the visitor wants ChatGPT-adjacent image integration. It is fine to address that term, but the page should still be honest about whether it documents a real project workflow or an official API product.

Imagen 2 API

Imagen 2 is a separate Google model family. It belongs in comparison discussions, not as if it were the same API under another name.

Best next step

Choose the path that matches what you actually need:

• Open GPT Image 2 if you want to test prompts and outputs first
• Read how to use GPT Image 2 if your main problem is workflow and prompting
• use the project routes above if you are already building around generation logic