GPT-4o’s Native Image Generation

Image by OpenAI

Just when I thought Google would hold the throne for a while as the best AI image editing model with the recently released Gemini 2.0 Flash, I was wrong. Today, OpenAI released GPT-4o with native image generation. This new model allows you to generate images, edit a single image with text prompts, and even combine multiple images into a single photo.

Unlike the previous image generator in ChatGPT powered by Dall-E 3, the new image generator is part of the GPT-4o model. Yes, GPT-4o is an “omnimodal” model capable of processing and generating text, audio, and images.

The shift from separate models to native integration within GPT-4o is a huge architectural advancement, enhancing performance and capabilities through tighter coupling of language understanding and visual synthesis.

Initial access to this new feature is rolling out to Plus, Pro, Team, and Free ChatGPT users starting in March 2025. Access for Enterprise and Education users, as well as API access for developers, is expected to follow soon.

If you want to learn more about how it works, check out the white paper here.

How to Access

There are few ways to try the new model:

1. ChatGPT: This is the easiest and the most straightforward way to try the new image generator/editor. Update your ChatGPT desktop app or access via chatgpt.com and describe the image you want to generate.
2. Sora: Notice that OpenAI added a brand new “Images” tab on the left panel of the website. You can remix or turn the image into video using Sora.

Image by Jim Clyde Monge

Image Generation Examples

Let’s start with image generation. Personally, I never used ChatGPT to create AI photos because Dall-E 3’s quality was poor, and the aspect ratio was always stuck at 1:1. However, with the recent GPT-4o update, the quality has improved significantly, and the aspect ratio can now be customized.

I tried it myself, and the results are just as impressive as the sample images.

Prompt: Generate a photorealistic…