GPT Image

basicwidespread

GPT Image is OpenAI's image generation model family, the successor to DALL-E: it draws images or transforms existing ones based on text prompts, and is also available in ChatGPT.

In plain terms

GPT Image is the 'artist' behind ChatGPT that creates an image from your description. It is like a fast-working graphic designer: you say what you want, and if you don't like the result, you ask for changes in words, such as 'make it evening light' or 'make the text red'.

You can create posters, illustrations, product photo sketches, diagrams, or invitations. You can also upload your own photo for it to transform or redraw in another style.

Its predecessor, DALL-E 3, was also available in ChatGPT, but there the chat program only forwarded the text request to a separate drawing model. In contrast, GPT Image is a system tightly integrated with the conversational model. Therefore, it better follows long, detailed requests, writes more readable captions on the image, and takes into account the earlier part of the conversation.

How it works

You describe what you want to see in text, and if needed, attach a sample image or a photo to be transformed.
The model processes text and image in a shared system. It breaks both down into small units called tokens.
According to OpenAI, the model generates the image step by step, similar to how a language model writes a sentence word by word. It pays attention to your request and the conversation history. The company has not disclosed the precise technical details.
Built-in filters check the request and the finished image. Provenance data (C2PA) stating that the image was made by AI is also embedded in the file, but this data can easily be lost, for example through a screenshot or re-saving.
If something is not right, you request a modification with another sentence. The model then redraws the existing image, so sometimes details you did not intend to change also alter.

An everyday example

You are preparing for your child's birthday party and type into ChatGPT: "Draw a dinosaur-themed invitation with the text: Bence is 7 years old, Saturday 3 PM." Within a minute or two, you receive a printable image, and if the name was misspelled, you can correct it with one sentence.

Who is affected

marketerYou can create campaign drafts, social media images, and captioned ad designs in minutes without waiting for a graphic designer. → Marketing

entrepreneur, executiveYou can get illustrations, product visualizations, or logo ideas at low cost, but legal and quality checks remain your responsibility.

developerYou can integrate image generation and editing into your own application or agent via API. → Code

everyoneYou are increasingly encountering images online created by such models, so it is worth viewing what you see critically.

A common misconception

MythGPT Image searches the internet and stitches together existing images.

In realityIt does not search or cut and paste: it creates every image anew from what it learned during training about the relationships between images and text. The style of the training images may still be reflected, which is why there is ongoing debate about what counts as copying.

Limits and risks

It may make mistakes in details: misspelled captions, inaccurate maps or diagrams, strange hands; always review the image before use.
Realistic fake images can be created with it, so a photo-like image is no longer proof of anything on its own.
The copyright situation is uncertain: depicting protected characters, the style of living creators, or real people carries legal and ethical risks.
Content filters may reject certain requests, and uploaded images are processed on the provider's servers, so do not upload sensitive photos without due consideration.

Similar, but not the same

DALL-E – DALL-E was OpenAI's previous image generator family, which in ChatGPT only processed text prompts passed through. Its successor, GPT Image, integrated with the conversational model, follows requests and captions more precisely.
ChatGPT – ChatGPT is the application where you converse; GPT Image is the model that generates images in the background.
Midjourney – Midjourney is another company's image generator, favored mainly for its artistic visual style. GPT Image's strength lies more in instruction following, text rendered in images, and conversational editing.

Related concepts

Broader

generative AI, image generation, multimodal model

Narrower

image editing via text instructions

On the edge

DALL-E, Midjourney, Stable Diffusion, deepfake, prompt

A brief history

OpenAI introduced the first DALL-E image generator in January 2021, followed by DALL-E 2 in 2022 and DALL-E 3 in 2023. In March 2025, a new image generator based on the GPT-4o model was added to ChatGPT. Developers made it available via API under the name gpt-image-1 from April 2025, and since then, further versions have appeared, such as GPT Image 1.5.

Where things stand

By the end of 2026, the GPT Image model family generates images in ChatGPT and is integrated via API into many other applications. It is one of the best-known image generators, in strong competition with models from Google and other companies.

Status: 2026 November

Also known as

GPT Image · gpt-image-1 · GPT Image 1.5 · GPT-4o image generation · 4o Image Generation · ChatGPT Images · ChatGPT image generator · OpenAI image model

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.