ChatGPT Images is OpenAI’s image generator, built into ChatGPT. It replaced DALL-E: the DALL-E 2 and DALL-E 3 API was retired on May 12, 2026, and the DALL-E GPT was removed from ChatGPT on August 30, 2026. If you followed an older DALL-E guide, this is the tool to use now.
The current version, ChatGPT Images 2.5, launched on September 8, 2026. OpenAI says it cuts generation time by up to 50% compared with Images 2.0 and holds on to reference images more faithfully when editing.
Ratings on this page are provisional until our hands-on test is published. Facts were checked against OpenAI’s documentation on the date shown above.
Who Can Use It
Every ChatGPT plan gets the same image model, including Free. Paid plans raise how many images you can make before ChatGPT asks you to wait. OpenAI doesn’t publish a fixed number for the free limit. “Images with thinking”, which plans the image before generating it, needs a Plus, Pro, or Business plan.
What It Does Well
Plain-Language Prompts
There’s no parameter syntax. Describe the image in sentences, then ask for changes the same way: “make the sky darker”, “move the text to the top”.
Editing
Select part of an image and describe the change, or upload a reference image and ask ChatGPT to keep its subject, style, or layout. You can also start from a sketch or a template.
Text in Images
Posters, signs, labels and packaging mockups are a strength. Spell out the exact words in quotes in your prompt.
Using It Through the API
Developers use the gpt-image-2 model (released April 21, 2026) through the Images API, which supports both generation and editing, including inpainting with image inputs. The API documentation lists:
| Setting | Options |
|---|---|
| Recommended sizes | 1024x1024, 1536x1024, 1024x1536 |
| Custom sizes | Multiples of 16, aspect ratio between 1:3 and 3:1, up to 3840 px on the long edge (above 2560x1440 is experimental) |
| Quality | low, medium, high, xhigh, max, auto |
| Transparent background | Yes, with PNG or WebP output |
| Output formats | PNG, JPEG, WebP |
Check OpenAI’s pricing page for current per-image costs.
Tips for Better Results
- Describe the scene, not keywords. “A fluffy orange tabby on a vintage windowsill in soft morning light” beats “cat, window, morning, photorealistic”.
- Iterate in the same chat. Follow-up edits keep the context of earlier images.
- Quote exact text. Put any words that must appear in the image in quotation marks.
- Name the format. Say “square”, “wide 16:9 banner” or “tall phone wallpaper” to set the aspect ratio.
Commercial Use
OpenAI’s Terms of Use assign you ownership of what you create, so you can generally use images commercially. Terms change, so check the current version before using images in paid work, and avoid prompts that copy trademarks or real people.