ChatGPT Images: Create Images from a Prompt
On this page
ChatGPT Images is the built-in image generator for ChatGPT. Describe what you want in plain language, and ChatGPT produces the image directly in the conversation, where you can then refine it. As of November 4, 2026, it runs on Images 2.6, which generates images faster and maintains details more consistently across edits, according to OpenAI.
What’s available and who can use it
According to the OpenAI Help Center, ChatGPT Images is available on all plans via the web, iOS, and Android, as well as in Codex. The “reasoning” version, which takes more time to plan the image, is reserved for Plus, Pro, and Business plans and is coming soon to Enterprise and Edu. Generation can take a few minutes depending on request complexity; you can continue using ChatGPT while it works.
| Reference | What to know |
|---|---|
| Start creating | Ask for an image in the conversation, or select Images from the menu |
| Format | Any aspect ratio, using the format selector or within the prompt |
| Text and background | Text inside the image, added details, transparent background on request |
| Motion | Looping GIFs, rolling out now |
| Storage | Every generated image is automatically saved in Images |
Your first generation
For example, type: “Generate an image of a Parisian café at sunset, photography style, with customers seated on the terrace.” The image appears in the thread, where you can immediately request modifications or variations without switching tools. To get started faster with common formats like posters or logos, the Images tab also offers customizable templates.
What makes a good prompt
Five pieces of information separate a rough image from one that matches your vision. The main subject answers “what?”: a cat, a landscape, a logo. The context or environment answers “where?”: on a desk, in a forest, on a white background. The visual style answers “how?”: photorealistic, watercolor illustration, flat design. The mood sets the atmosphere: warm, dramatic, minimalist, or cheerful. Finally, technical details frame the shot: close-up, aerial view, portrait.
Combine these five decisions into a prompt like this: “A red fox sitting on a mossy rock in a coniferous forest at sunrise. Wildlife photography style, soft golden light, shallow depth of field, vertical framing.” Compare this to “make me a nice fox image”: the vague prompt makes no decisions, so the model makes them all for you.
Text in images
The Help Center states that ChatGPT Images can follow instructions to add text within an image. Request, “Create a blog banner with the text ‘Welcome to my site’ in elegant typography on a blue and purple gradient background,” placing the exact text in quotes. Always proofread the spelling before publishing: a distorted letter can easily go unnoticed.
This capability opens up use cases where text is part of the design: marketing visuals, covers, poster mockups.
Iterate instead of saying everything at once
After the first generation, refinement happens in the same conversation: “make the background darker,” “add a hat to the character,” “keep the same composition but switch to black and white.” This is also the remedy for the most common mistake: stuffing two hundred words into a single prompt. Proceed in three steps: a base scene, then elements to add, then the style. You’ll see at each step what your adjustment actually produced.
Strengths and limitations
The model handles detailed instructions, text, conversational consistency, and a wide variety of styles with ease. According to OpenAI, Images 2.6 also improves consistency for characters, products, and brand details across edits; for recurring subjects, you can save a reference photo and call it later using @ followed by its name.
Know its limitations before committing to a client deliverable. Highly technical requests, such as engineering diagrams or dimensioned plans, yield approximate results. Some requests are refused under usage policies. And a generated image illustrates an idea: it does not document a real object or event.
Your turn
Three iterations on the same scene. Request a first image using a five-element prompt. Then modify only one parameter at a time: the mood, then the format, then the style. Note what changed and what stayed stable at each step, and save the version you prefer from the Images tab.
Official sources: Images in ChatGPT (OpenAI Help Center) · Introducing Images 2.6 (OpenAI, November 4, 2026)
Key takeaways
- ChatGPT Images, now on version 2.6, is available on all plans; the reasoning version is reserved for Plus, Pro, and Business
- A good prompt defines the subject, context, style, mood, and framing
- All formats are possible, via the selector or in the prompt
- Iterate with small tweaks rather than overloading a single prompt
- Images are saved in the Images tab; proofread text and verify intended use before publishing