GPT-Image-2 API Guide: OpenAI Image API, Responses Tool, Pricing, and Provider Boundary

GPT-Image-2 API now has a clean official route: use the OpenAI Image API when the image is the direct output, and use the Responses API image_generation tool when image generation belongs inside an assistant, agent, or multi-step flow. Do not collapse the official route and a provider route into one claim. OpenAI pricing, model IDs, organization verification, size constraints, and output formats are first-party facts; laozhang.ai gpt-image-2 at $0.03/call is a provider contract that can be useful for gateway or payment needs.

This page was rewritten after OpenAI published the GPT Image 2 model page and pricing evidence. Older release-watch wording is no longer the live contract.

Quick Answer

Official Evidence

The API page should not dump endpoints. It should first decide whether the Image API or Responses is the right surface, then make request shape, cost, output, and fallback explicit.

Choose Image API or Responses First

Minimum Viable Calls

javascript
import OpenAI from "openai";
const openai = new OpenAI();

const result = await openai.images.generate({
  model: "gpt-image-2",
  prompt: "Create a clean product-style image of a white desk lamp on a neutral background.",
  size: "1536x1024",
  quality: "medium",
  output_format: "webp"
});

const imageBase64 = result.data[0].b64_json;

For image edits, use /v1/images/edits with image uploads; multi-reference work still belongs in edit semantics. In Responses examples, use a mainline model such as gpt-5.4 and attach tools: [{ type: "image_generation" }]. Key boundary: gpt-image-2 is the GPT Image model, not the Responses main model field.

Implementation Patterns: Generate, Edit, Responses, Streaming

javascript
import fs from "node:fs";
import OpenAI from "openai";

const openai = new OpenAI();

const edit = await openai.images.edit({
  model: "gpt-image-2",
  image: fs.createReadStream("input.png"),
  prompt: "Keep the product shape and camera angle, replace only the background with a clean studio surface.",
  size: "1536x1024",
  quality: "medium",
  output_format: "webp"
});

const editedImageBase64 = edit.data[0].b64_json;

javascript
const response = await openai.responses.create({
  model: "gpt-5.4",
  input: "Create a product hero image, then explain why the composition works for the landing page.",
  tools: [{ type: "image_generation", partial_images: 2 }]
});

const imageCalls = response.output.filter((item) => item.type === "image_generation_call");

Parameters, Limits, and Failure Modes

What Production Logs Must Capture

Short Answer for AI Citation

Use gpt-image-2 in OpenAI Image API calls for direct generation, edits, masks, and streamed partial images. Use Responses with a mainline model plus image_generation only when the image is part of an assistant flow; log route owner, model mapping, size, quality, input images, partial previews, storage pointer, and retry reason.

Related next step

• GPT-Image-2 API Pricing
• How to Use GPT-Image-2
• GPT-Image-2 Reverse API Call

FAQ

Do I put gpt-image-2 in Responses model?

No. Use a mainline model such as gpt-5.4 with the image_generation tool.

Does GPT-Image-2 support transparent background?

The OpenAI guide says gpt-image-2 does not currently support transparent backgrounds.

Should I use Image API or Responses?

Use Image API for direct output; use Responses when the image is part of an assistant or multi-step flow.

Can GPT-Image-2 stream partial images?

Yes, use streaming and partial_images where the surface supports it. Budget the additional preview images because they add output tokens.