How to create and edit images in ChatGPT, now that DALL·E is gone

Where ChatGPT image generation stands now

As of October 2026, image generation in ChatGPT is a feature called ChatGPT Images. It runs on ChatGPT Images 2.5, a model released on September 8, 2026. It is available on every plan, including Free, on the web, iOS and Android.

When we wrote about this in April 2025, image generation with GPT-4o had just launched, and the older DALL·E 3 was still around. Here is what happened after that.

HIKARI'S TAKEAny image you make in ChatGPT today comes from Images 2.5. You can't pick DALL·E anymore, so there's no longer a choice to agonize over.

How to create images in ChatGPT

The basics are the same as in 2025: describe the image you want. What's new is that there are more ways to start. Here are the steps, based on OpenAI's help center (checked October 26, 2026).

1. Ask in the chat.Type something like "Make an image of…" in the usual message box, and ChatGPT creates it.
2. Start from Images.Open "Images" from the sidebar or "More," and enter a prompt. Everything you've made so far is collected there too.
3. Use a template.Go to Images, then Templates, pick a starting point such as a poster or logo, and fill in the details. If ChatGPT asks follow-up questions, choose an option or type your own answer.
4. Start from a sketch (mobile).In the app, type "@" in the message box, choose Sketch, draw, and then describe the image you want.
5. Set the aspect ratio.Use the aspect ratio picker, or write something like "16:9" in your prompt.

Depending on the request, an image can take a few minutes. You can keep using ChatGPT while you wait.

How to write prompts that get you closer

Based on OpenAI's help center and announcements, here is what helps most for work images.

• Say what the image is for.For example, "a letter-size flyer for the shop window" or "a square Instagram post." The help center also suggests stating the purpose and the exact wording for posters.
• Put any text you want in quotation marks, word for word.OpenAI says text rendering improved a lot with Images 2.0. Long sentences and small type can still come out wrong, so zoom in and check the result.
• If you want a transparent background, say so.ChatGPT supports it.
• If you have a reference photo, upload it first.OpenAI says Images 2.5 keeps the subject of a reference photo more faithfully.

HIKARI'S TAKEStart by simply asking in the chat. Once you're comfortable, try templates and aspect ratios. For images with text, zoom in and check every word.

How to edit images in ChatGPT

You can edit both images you made in ChatGPT and photos you took yourself. There are two main ways.

Edit by describing the change

Select an image or upload a photo, then describe what you want changed. Naming the spot, as in "make the background white" or "remove the logo in the top right," makes it more likely to come out the way you want. OpenAI says Images 2.5 holds on to earlier changes better across several rounds of edits.

Select an area to edit (the editor)

Select an image in a conversation or in Images in the sidebar, and the editor opens. Here is what it can do.

Edit on your phone

In the app, tap an image to open it full screen. Tap Edit to describe a change, or tap Select to brush over an area. Use the slider to change the brush size, tap Next, and describe the change. On mobile you also get Resize and Remove tools, and you can leave comments directly on the image to point out what should change.

HIKARI'S TAKEUse the chat for small fixes and the editor when you want to change a specific spot. And save the original to your own device first, just in case.

ChatGPT image generation: free and paid plans

You can make images on the Free plan. What changes by plan is how many you can make, and whether you get the feature that plans the layout before generating. The table uses the wording on OpenAI's US pricing page and the US prices from OpenAI's help center and announcements (checked October 26, 2026).

Images with thinking, which plans the layout before it generates, is available on Plus, Pro and Business, and OpenAI says Enterprise and Edu are coming soon. When it announced Images 2.0, OpenAI said this feature can produce up to eight related images from a single request.

Go is OpenAI's low-cost plan. On January 16, 2026, it became available everywhere ChatGPT is offered, including the US, where it costs $8 a month. Ads appear only on the Free and Go plans; Plus and Pro stay ad-free. Prices can change, so check the pricing page before you subscribe.

HIKARI'S TAKETry the Free plan first, and think about Go or Plus once you're making a lot of images for work. Prices change often, so take a look at the pricing page before you sign up.

Commercial use and copyright for images made in ChatGPT

Our 2025 article said almost nothing about rights. If you use these images for work, this is the part that matters most.

Who owns the images you make

OpenAI's Terms of Use (effective January 1, 2026) say that, to the extent the law allows, you keep ownership of your input and you own the output, including the images you make. The same terms also say that, because of how AI works, other users may receive similar output. Owning an image and that image being unlike anything else are two different things.

Owning an image also does not mean it is protected by copyright. In its January 2025 report on copyrightability, the U.S. Copyright Office concluded that prompts alone do not make an AI output protectable. Protection can cover what a human contributes, such as a human-made work that is visible in the output, or creative arrangement or changes to it.

Don't use images that resemble existing works or real people

As a general rule of U.S. copyright law, an image that copies protected parts of an existing work can infringe on it, whether a person or an AI made it. This is general information, not legal advice. For images of real people, OpenAI says it has tightened what can be created.

Images record that they were made with AI

Images made in ChatGPT carry two signals: metadata that shows where they came from (C2PA) and an invisible watermark (SynthID). You can upload an image at openai.com/verify to check whether it was made with OpenAI's tools. OpenAI also notes that these signals do not guarantee that the content is accurate or that anyone holds the rights to it.

If you don't want your photos or prompts used to train OpenAI's models, you can turn that off in settings on any plan.

1. Don't steer toward existing characters, logos or a specific artist's style.If the result ends up looking like something else, don't use it.
2. Use photos of real people only with their permission.That includes your employees and customers.
3. Have a person check text, numbers and maps in the image.A polished look does not mean the content is right.
4. Have a second person review anything public, such as ads or packaging.Don't leave the call to the person who made it.
5. Save the images you use to your own drive.To remove an image from Images in ChatGPT, you have to delete the whole conversation that created it.

HIKARI'S TAKEThe images you make are yours, but that doesn't make it fine to use one that looks like someone else's work. Run through the five points above before anything goes public.

Checking our 2025 article

In April 2025, right after GPT-4o image generation launched, we published a guide to this feature in Japanese. Here is how what we wrote then holds up today.

Our biggest miss was saying nothing about rights when you use these images for work.

The original article said the images were handy for social media and stickers, but never covered who holds the rights or what to do if a result looks like something else. Around the same time, another of our articles (also in Japanese) explained how to make images in the style of a famous animation studio. Reading it again now, rights were the point that mattered most to users.

HIKARI'S TAKEThe how-to mostly still works. What changed was DALL·E, and what was missing was rights. This article fills that gap.