The fastest way to make images with ChatGPT is to type a description into a normal chat, such as "Create an image of a lighthouse at dusk in watercolor", and wait for the picture. ChatGPT Images 2.5, which OpenAI released on September 8, 2026, added templates, a Sketch tool on phones and comments you can place directly on an image. The sections that follow give the steps for each way in, editing and saving, copyable prompts, and the limits on each plan.
4 ways to make images with ChatGPT
Every ChatGPT plan, Free included, can make pictures on the web, in the desktop app and in the phone apps. Here are the four ways to start one, as of September 2026:
| Method | Works on | Cost | Best for |
|---|---|---|---|
| Ask in any chat | Web, desktop, iOS and Android | Every plan, within its image limits | A quick picture while you're already chatting |
| Templates | The Images page (not yet in Work mode) | Every plan | Posters and merch designs from a starting point |
| Sketch | ChatGPT mobile app | Every plan | Layouts that are easier to draw than to describe |
| Upload a photo | Web and apps | Every plan | Restyling or editing a picture you already have |
How to create an image with ChatGPT in a chat
You don't need a special mode to use ChatGPT to make images: ask for the picture in any conversation, or open the Images page.
- Open ChatGPT in a browser or in the mobile or desktop app.
- Start your message with "Create an image of", or pick More, then Images, and describe the picture there.
- Set the shape in the aspect ratio picker, or add it to your text: "16:9" for a wide banner, "9:16" for a phone screen.
- Give it time. OpenAI says a complicated picture may need a few minutes, and the chat stays usable in the meantime.
- Click or tap the finished picture to open it in the editor.
Save or download the image
ChatGPT files every picture you make on the Images page automatically, and the Create Image button there starts a new one. Open a picture and choose Copy for the clipboard, Save to put a copy on your device, or Share to send it to another app. Deleting a picture means deleting the chat that made it: on the web, open the three-dot menu beside that chat and choose Delete; in the app, press and hold the chat title and tap Delete.
Start from a template
Templates are starting points for popular formats such as a poster or merch, so you only add your details.
- Choose Images in the sidebar.
- Pick Templates, then a category, then the template you want.
- Fill in your details, such as the event name, the colors or the product, and send it.
- If ChatGPT asks about the style or other details, answer or pick an option, and add a reference image if it offers an upload.
Templates aren't available in Work mode yet. Read every word the template produced, because you only supplied part of the text.
Draw it with Sketch on your phone
Sketch lets you draw a rough layout in the ChatGPT mobile app and send it with written instructions.
- In the message box, enter @ and pick Sketch.
- Draw the rough shapes, switching colors, erasing or undoing strokes as you go.
- Tap the checkmark when the drawing is done.
- Add instructions and send them, for example "Turn this into a flat icon of a coffee cup with steam, navy and orange, on a white background."
How to use the ChatGPT image generator to edit a result
Edit a ChatGPT image by selecting the part you want changed and describing the change in the chat.
- Open the picture in the editor.
- Choose Select and highlight the area to change.
- Describe the change and name what must stay, for example "Replace the red umbrella with a yellow one. Keep her face, the street and the light the same."
- To change the shape, choose Aspect ratio. ChatGPT regenerates the image in the new ratio.
OpenAI warns that the highlight isn't exact, and an edit may spill outside it. Request a single change each time, which OpenAI recommends to keep the rest of the image from drifting, and compare each version with the one before. In the phone app, you can also open an image full screen and leave comments on the spots you want changed. For editing your own photos, including photos of people, read can ChatGPT edit photos.
How to write a prompt ChatGPT follows
OpenAI's guidance is that one to three clear sentences usually do the job. Cover what decides whether the image works:
- Purpose. Who it's for or where it goes: "for a bakery's menu board".
- Subject and action. "A golden retriever catching a frisbee" beats "a dog".
- Setting and composition. Where it happens, and what sits in the foreground, background, left and right.
- Style. Photo, watercolor, flat vector, pencil sketch.
- Light as a direction. "Low morning sun coming through a window at the right" instead of "beautiful lighting", which OpenAI gives as an example of wording that's too vague.
- Constraints. Things to leave out, such as logos or extra text.
- Exact words in quotes. Type the capitals the way they should appear, spell rare names out one letter at a time, and say whether any other words are allowed.
The prompts below follow that shape. They work as written in ChatGPT and in other image tools.
Realistic images
To make realistic images with ChatGPT, describe the shot like a camera setup: how far away the camera is, the lens, the light and the time of day.
A photorealistic close-up of a ceramic mug of black coffee on a scuffed oak cafe counter, photographed at counter height with an 85mm lens. Soft early sun through a window at the right, shallow depth of field, a blurred street outside. Natural colors, no text, no logos.If the result still looks artificial, the guide to making AI images look real covers skin, hands and lighting fixes.
Graphics with a few words
OpenAI's advice for words in an image is to keep them brief and exact, and to give the font style, size, color and position. Its help center says ChatGPT can also make an image's background transparent, which is handy for a logo or a sticker.
A square poster for a neighborhood book swap. At the top, a bold headline in large dark green letters: "Book Swap". Below it, smaller text: "Bring a book, take a book". A flat illustration of three paperbacks piled beside a potted fern on a cream background. No other text.Posters that carry a date, a price or an address need a slower proofread, and the reasons AI misspells text in images explain what to look for before you print. For event details, can ChatGPT make a flyer covers what to proofread.
Illustrations with room for a headline
A flat vector-style illustration of a red bicycle leaning against a yellow wall under a window box of daisies. Simple shapes, thick outlines, three colors plus white, in a wide 16:9 frame with the right third left blank for a headline.Several reference photos
When you attach more than one image, say which is which. OpenAI suggests numbering them by order and using words like left, right, foreground and background, for example: "Image 1 is my living room. Image 2 is the sofa I want. Put the sofa from image 2 against the back wall of image 1, and keep the rug, lamp and window as they are." The AI image prompt library has more starting points by category.
How to make images with ChatGPT for free
You can make images with ChatGPT for free on the Free plan, up to a usage limit. OpenAI's Free Tier FAQ says image creation has its own allowance, separate from chat, and that it can be stricter than on paid plans. ChatGPT shows a notice when the allowance runs out, and you have to wait before making more. Moving up to a Plus, Pro or Business plan resets that usage immediately.
How ChatGPT's pricing page describes image creation on each plan (prices in US dollars, as of September 2026):
| ChatGPT plan | Monthly cost | Image creation |
|---|---|---|
| Free | $0 | Limited and slower |
| Go | $8 per month | More than Free |
| Plus | $20 per month | More complex and accurate, including Images with thinking |
| Pro | From $100 per month | Unlimited and faster, subject to abuse guardrails |
Images with thinking is also on Business plans. OpenAI says the limits can change over time. The guide to ChatGPT's image cap explains resets and what to do in the meantime.
Where ChatGPT images fall short
A few jobs still need a careful check, or a second tool, after ChatGPT makes the picture.
- Dense text. For dense copy or print-critical typography, OpenAI's own advice is to proofread each word and complete the layout in a design program.
- Imprecise selections. A highlighted edit can change pixels outside the highlight. Look at faces, hands and edges after each edit.
- Pixel sizes. The aspect ratio picker sets the shape, not the pixel count. If a print shop or a platform needs exact pixels, check the downloaded file's dimensions before you send it.
- Real people and brands. OpenAI asks you to have permission before depicting a real person, and to ask for an original design rather than a copy of a brand, product or artwork.
- Provenance. OpenAI marks ChatGPT Images 2.5 pictures two ways: C2PA metadata and Google DeepMind's invisible SynthID watermark. ChatGPT image watermarks explains what that means for you.
- Old guides. Instructions that tell you to pick DALL·E are out of date. OpenAI retired its DALL·E GPT on August 30, 2026, and points users to ChatGPT Images instead.
Troubleshooting ChatGPT images
| Problem | Likely cause | Fix |
|---|---|---|
| ChatGPT says you've reached a limit | You've hit your plan's image allowance | Wait and try later, or upgrade, which resets your usage |
| A word is misspelled | The text is long or unusual | Shorten it, put it in quotes, and spell uncommon names letter by letter |
| The edit changed things you didn't select | Selections aren't exact | Name everything that should stay untouched, and change one thing per request |
| The image is the wrong shape | No ratio was given | Use the aspect ratio picker, or choose Aspect ratio in the editor |
| The request is refused | It conflicts with OpenAI's usage policies | Rephrase around an original subject, and use a real person's likeness only with their permission |
| Details drift after several edits | Many changes stacked in one chat | Start from the version you liked and restate what stays fixed |
| An old image is missing | The chat it came from was deleted | Download the pictures you want to keep |
Make the same image in Image Generator
Image Generator in Toybox AI takes the same kind of prompt in a single form, with five set shapes and room for reference photos, and no chat thread to manage.
- Sign in to Toybox and open Image Generator. A picture currently costs 50 credits, and a new account's free starting balance is 20 credits, so buy a top-up or pick a plan first.
- Paste your prompt into "Describe your image".
- Optional: tap "Add reference photos" and attach up to 3 pictures to steer the style, subject or layout.
- Under "Size", pick Square (1:1), Portrait (3:4), Landscape (4:3), Story (9:16) or Wide (16:9).
- For a picture with words, turn on "Enhanced quality" under "More options" (currently +45 credits).
- Tap "Create image" and allow about 30 seconds.
- Check every word and face. With a Pro plan, describe a fix in "Want to change something?" (currently 85 credits), or open the result in Smart Editor, which also needs Pro.
Image Generator has no resolution setting, so check the downloaded PNG's pixel size before you print it. A face in a reference photo is a guide, not a copy, so expect some differences. Plan and top-up prices are on the pricing page, and AI image generators compared with ChatGPT weighs the other options.