ChatGPT Image Generator: How It Works in 2026 (Guide)
The ChatGPT image generator is the tool inside ChatGPT that turns a written prompt into a finished image, or edits a picture you upload. It is powered by GPT Image 2.5, OpenAI's latest image model, released on 8 September 2026. You can make posters, product shots, illustrations and photo edits, and it is available on every plan, including free.
The ChatGPT image generator lets you create and edit pictures just by describing what you want in plain English. Type a request such as "a watercolour poster of a London street at dusk" and ChatGPT paints it in seconds. You can then ask for changes in the same conversation, the same way you would ask for a rewrite of a paragraph.
The tool has changed enormously this year. OpenAI retired DALL-E 3, the model that powered it for years, in May 2026 and replaced it with its own GPT Image family. The current model, GPT Image 2.5, arrived on 8 September 2026 with faster generation, sharper editing and new tools such as Sketch. OpenAI says people now create more than 3 billion images every week across ChatGPT and its image models.
This guide explains how the generator works, what you can create with it, how to use it step by step, and what the limits are on each plan.
In this guide:
For most of ChatGPT's life, image generation was powered by DALL-E 3, a dedicated image model OpenAI first released in 2023. That ended in 2026: OpenAI retired DALL-E 3 on 12 May 2026, both inside ChatGPT and on its developer API, and the separate DALL-E tool in ChatGPT was switched off at the end of August 2026.
Every image you request in a normal chat is now generated by the GPT Image family. GPT Image 2.0 arrived in April 2026 with better text rendering and multilingual support, and GPT Image 2.5 launched on 8 September 2026 as the current model.
OpenAI says Images 2.5 produces more natural lighting and richer textures, keeps the people in your reference photos more recognisable, and generates images up to 50% faster than version 2.0. If you are new to all of this, our guide to whether ChatGPT can create images covers the basics.
The generator works like the rest of ChatGPT: you describe what you want in ordinary language and the model produces it. Under the bonnet, GPT Image 2.5 is a text-to-image model. It reads your prompt, works out what you are asking for, and paints the result, guided by the enormous number of pictures and captions it learned from during training.
What sets it apart from older image tools is the conversation around the image. You can upload your own photo and ask for changes: place the subject in a different setting, turn it into an illustration, or restyle it for a party invitation. Images 2.5 is better at editing only what you mention and keeping the rest of the picture consistent across edit after edit.
It also handles complex briefs more reliably than earlier models. It can follow detailed layout instructions, produce images with transparent backgrounds, and hold a consistent style across a whole series, which matters when you need a matching set of graphics for a project.
Almost any still image you can describe. People use the generator for:
Illustrations and artwork, from watercolour scenes to logo concepts
Posters, flyers and social media graphics with readable text
Product photos and mockups for small shops and marketplaces
Edits of your own photos, such as new backgrounds, restyling or added objects
Storyboards, charts and infographics for presentations
Fun transformations of family photos, such as the viral 1980s portrait prompt
If you want to change an existing picture rather than create one from scratch, our guide to ChatGPT photo editing walks through that workflow in detail.
Generating an image takes three steps and no setup:
Open ChatGPT on the web, the desktop app or the mobile app and start a new chat.
Type your request in plain English, for example: "Create an image of a fox reading a newspaper on the morning commute, in the style of a New Yorker cartoon."
Press send and wait a few seconds. The image appears in the chat.
To edit, keep talking in the same chat. Ask for a different colour, a new background or clearer text and the model updates the picture. You can also attach a photo first, then describe what you want changed.
Images 2.5 introduced two tools that make starting easier. Sketch lets you draw a rough doodle directly in ChatGPT and use it as a guide for the final image. Type @Sketch, scribble the layout of a room or the outline of an outfit, add a short description of the style, and ChatGPT turns the doodle into a finished picture. You do not need to be able to draw.
Templates give you a head start on popular formats. Instead of describing a poster or a piece of merchandise from scratch, you choose a template such as Poster or Merch and fill in the details. Both features are available to every user, on every plan.
In ChatGPT itself, image generation is included on every plan, including the free tier. OpenAI does not publish fixed daily allowances, and the limits move with demand. As a rough guide, free users can generate a handful of images a day, while Plus subscribers commonly report around 100 or more. If you reach your limit, it usually resets within a few hours. Our comparison of ChatGPT plans explains what each subscription includes.
Developers building image generation into their own apps pay per token through the OpenAI API. Images 2.5 is available there as GPT-Image-2.5 Flare, the fast default, and GPT-Image-2.5 Sunburst, which trades longer generation times for extra precision.
The generator is impressive, but it helps to know where it still struggles:
Small text and dense typography can come out garbled, although version 2.5 improved this
Large layout changes, such as moving several objects at once, can introduce artefacts elsewhere in the picture
Briefs with many specific requirements may need a few attempts to get right
Requests that break OpenAI's content policies are refused, including misleading images of real public figures
Every image carries an invisible watermark and C2PA metadata that identify it as AI-generated
If you plan to use generated images commercially, OpenAI's terms allow it, but disclosure rules and platform policies vary, so check the ones that apply to you.
ChatGPT's image generator is built into every chat and included on every plan, including free
It runs on GPT Image 2.5, released 8 September 2026, which replaced the retired DALL-E 3
You create images by describing them in plain English and refine them by chatting
New Sketch and templates features make posters, merchandise and layouts easier to start
Allowances vary by plan and demand, and paid plans allow far more images per day than free
Is the ChatGPT image generator free to use?
Yes. Image generation is included on every ChatGPT plan, including the free tier. Free users get a small daily allowance and paid plans such as Plus and Pro allow far more images per day. OpenAI does not charge separately for generating images inside ChatGPT.
Does ChatGPT still use DALL-E?
No. OpenAI retired DALL-E 3 on 12 May 2026 and switched off the DALL-E tool inside ChatGPT at the end of August 2026. All images are now generated by the GPT Image family, currently GPT Image 2.5, which OpenAI says is faster and better at editing.
How many images can I create per day?
There is no published fixed number, and allowances vary with demand. Free users typically manage a few images a day, while Plus subscribers commonly report around 100 or more. If you reach your limit, it usually resets within a few hours.
Can ChatGPT edit my own photos?
Yes. Upload a photo in a chat and describe the change you want, such as a new background or a different style. Images 2.5 is noticeably better than earlier versions at editing only what you ask for and keeping the rest of the picture intact across several rounds of edits.
Can I use images from ChatGPT commercially?
Yes, OpenAI's terms allow you to use the images you create, including commercially. Remember that every image carries an invisible watermark and C2PA metadata marking it as AI-generated, and that you are responsible for making sure it does not infringe anyone else's rights.
Why does ChatGPT refuse some image requests?
The generator applies OpenAI's content policy. It declines requests involving real public figures in misleading situations, explicit content, or anything that breaks its safety rules. Rephrasing a request without changing its intent will not lift a refusal.
The ChatGPT image generator has quietly become one of the most used creative tools in the world, with more than 3 billion images made every week. It is included on every plan, it understands plain English, and with GPT Image 2.5 it edits as well as it draws.
Try one simple prompt today: a poster, a product shot or a restyle of a photo from your phone. Then explore more ChatGPT guides on whatischatgpt.co.uk.
