ChatGPT Image Generator: How It Works in 2026

The ChatGPT image generator changed more in the past year than most people noticed. If you last made a picture with it back in 2024, you were talking to DALL-E, a separate model that ChatGPT called on like a plugin. That era is finished. OpenAI retired DALL-E 2 and DALL-E 3 on May 12, 2026, and image creation now runs inside the main model itself.

You still type a prompt into the same chat box. What happens behind it is different, the output is noticeably better, and the rules around using those images have real edges worth knowing before you put one on a client invoice. Here is how it actually works now, what the free tier gets you, where it wins, and when you should reach for something else.

What powers ChatGPT image generation now

DALL-E was a diffusion model bolted onto ChatGPT. You wrote a description, ChatGPT passed it to DALL-E, and DALL-E painted a picture in near isolation from the conversation. It was good for its time and famously bad at two things: putting readable text into an image and following a long list of specific instructions.

The replacement is OpenAI's GPT Image family, the same natively multimodal image generation that first showed up in the GPT-4o era in early 2025. The key word is native. Image generation is now part of the model instead of a tool it hands off to. Because the same model that reads your prompt also draws the picture, it holds far more context: the details you mentioned three messages ago, the correction you just made, the style you asked it to keep. Instead of a diffusion pass, it builds the image in a more sequential way, which is slower per image but much better at getting small details right. The most visible payoff is text. Signs, labels, posters, and mockups now come out with legible words most of the time, which was close to impossible with DALL-E.

If you want the wider picture of how the chat side of the model has evolved, our guide on how to use ChatGPT covers the fundamentals that make image prompting work better.

How to use it

You generate an image by describing it in plain language. No menus, no seed numbers, no separate app. Type "a warm, sunlit ceramic coffee cup on a wooden desk, soft morning light, shallow depth of field" and you get it back in the chat within a minute or so.

The real advantage over standalone tools is iteration. You do not restart from a blank prompt when something is off. You say "make the mug blue," "add steam," "same scene but from above," and the model edits its own previous output while keeping everything else consistent. That conversational back-and-forth is the feature people underrate most.

A few things it handles well:

• Text in images. Ask for a poster that says a specific headline, a shop sign, a quote card, or a labeled diagram, and the words usually render correctly. This alone makes it useful for social graphics and simple marketing mockups.
• Editing your own uploads. Drop in a photo and ask it to change the background, remove an object, or restyle the whole thing. You describe the change in words instead of masking areas by hand.
• Style control. It follows style cues well: "flat vector illustration," "1970s film photo," "isometric 3D," "watercolor." Reference a look and it will hold it across a series so a set of images feels like it belongs together.
• Aspect ratios and layout. Ask for a wide banner, a square post, or a vertical story and it composes for that shape rather than cropping a square.

The prompting rule that matters: be specific about the things you care about and vague about the rest. Naming the subject, the lighting, the mood, and the format gets you most of the way there. Overloading a single prompt with fifteen competing demands is where it starts to drop details.

Free vs Plus: what you actually get

You can generate images on the free tier of ChatGPT. The catch is volume. Free users get a small number of images per day, and the exact ceiling floats depending on server load, so on a busy afternoon you might hit the wall after a handful of generations and get asked to wait or upgrade.

ChatGPT Plus at $20/month raises that limit substantially, gives you priority during peak hours, and keeps you on the strongest current model. OpenAI does not publish a precise image cap for Plus, but in practice it is high enough that most creators and marketers never think about it during a normal workday. There is also a cheaper ChatGPT Go tier at $8/month that sits between the two, and ChatGPT Pro at $200/month that removes the practical limits entirely and is aimed at heavy daily users.

For almost everyone reading this, the honest answer is: start free, and upgrade to Plus only once you actually hit the daily limit often enough to be annoyed. The image quality itself is the same across tiers. What you pay for is throughput, speed, and not getting throttled.

Where it is strong

The generator earns its place for a specific profile of work. It is the best option when you want a usable image fast without leaving your chat, when the image needs real words in it, and when you expect to revise it a few times. Blog headers, social posts, quick product mockups, slide illustrations, thumbnail concepts, and first-draft visuals all fit here. It is also the most beginner-friendly generator on the market because there is nothing to learn beyond describing what you see in your head.

The conversational editing is the quiet superpower. Being able to say "keep everything, just swap the season to winter" and get exactly that back, without re-rolling the whole image, saves more time than any single prompt trick.

Where it falls short

Being honest here matters more than being enthusiastic.

It is not the top choice for fine-art quality or a strong, opinionated aesthetic. Outputs can look competent and slightly generic, the visual equivalent of a well-written but forgettable email. You get less granular control than dedicated tools offer: no precise seed control, limited fine-tuning, and no deep parameter tweaking. It can still fumble complex scenes with many interacting elements, hands and crowds remain occasionally awkward, and very long text passages break down even though short text is now reliable. And because generation is sequential rather than instant, it is slower per image than a pure diffusion tool, which matters if you need to produce fifty variations quickly.

Commercial use: what you can and cannot do

Under OpenAI's terms of use, you own the images you generate and you are allowed to use them commercially: marketing, ads, product design, merch, client work. That is the good news, and it removes a real worry for freelancers and small teams.

Two caveats you should not ignore. First, ownership from OpenAI does not mean the image is free of every legal risk. If you prompt it to closely copy an existing character, logo, celebrity likeness, or branded style, you can still land in copyright or trademark trouble no matter what the terms say. The permission is to use what you made, not to launder someone else's IP. Second, in the United States, purely AI-generated images generally cannot be copyrighted by you, which means a competitor could reuse the same output. If exclusivity matters, add meaningful human editing on top.

One more practical detail: every image ChatGPT produces now carries C2PA content credentials, invisible metadata that flags it as AI-generated. It is good for transparency, but be aware some platforms and clients can read it, so an AI image will not always pass as a photograph.

When a dedicated tool is the better call

ChatGPT is the flexible generalist. For certain jobs, a specialist beats it clearly:

• Midjourney wins on pure artistic quality. If you want striking, art-directed, editorial-grade images with a distinct aesthetic, it is still ahead. It is weaker at spelling out text, so pair it accordingly.
• Ideogram is built around text-in-image and typography. For logos, posters, and design-heavy layouts where the words are the point, its accuracy leads the field.
• Adobe Firefly is the safe pick for commercial work with legal peace of mind. It is trained on licensed content and Adobe Stock, ships with commercial indemnification on eligible plans, and lives inside Photoshop, which matters if your workflow is already in Adobe.

A useful setup for most creators is ChatGPT for fast, iterative, text-in-image work, plus one specialist for the jobs where quality or safety is non-negotiable. To compare the full field side by side, see our roundup of the best AI image generators, and if you are choosing which assistant to build your workflow around, our best AI chatbots comparison is worth a read.

FAQ

Is ChatGPT still using DALL-E for images?

No. OpenAI retired DALL-E 2 and DALL-E 3 on May 12, 2026. Image generation now runs on OpenAI's newer GPT Image models, built directly into the main ChatGPT model instead of being a separate tool it calls. You use it the same way, by describing what you want in the chat.

Can I generate images with the free version of ChatGPT?

Yes. The free tier lets you create images, but with a low daily limit that can tighten when servers are busy. ChatGPT Plus at $20 per month raises the cap sharply and gives you priority speed. Image quality is the same on both tiers, so upgrade only when the free limit gets in your way.

Can I use ChatGPT images commercially?

Yes. OpenAI's terms grant you ownership of the images you generate and permit commercial use. Just avoid closely copying existing characters, logos, or brands, and remember that in the US a purely AI-made image usually cannot be copyrighted by you, so add human editing if you need exclusivity.

Which is better for text in images, ChatGPT or a design tool?

ChatGPT is now reliable for short text like headlines, signs, and quote cards, a big jump over old DALL-E. For typography-heavy work such as logos and detailed posters, a specialist like Ideogram still renders text more accurately and gives you more layout control.

AIpresso covers the AI tools worth your time in one short read a day. Join the creators and marketers who start their morning with the updates that actually matter.