Generate images with ChatGPT Images

Create visuals directly in the chat

ChatGPT Images is the built-in image generator for ChatGPT. Without leaving the chat or opening editing software, you can create illustrations, logos, posters, presentation slides, and social media visuals, or modify existing images. Version 2.6, released in November 2026, generates images faster and preserves details more consistently across iterations. The feature is available on all plans via web, iOS, and Android. Creation limits increase with your plan, ranging from “limited” on Free to “unlimited” on Pro, subject to abuse safeguards. On Plus, Pro, and Business plans, images can also leverage the model’s reasoning capabilities.

You can request an image just like any other message, or via the Images interface:

“Generate an image of a modern open-plan office with green plants and natural light streaming through large bay windows, professional photography style.”

Generation may take a few minutes depending on complexity; you can continue using ChatGPT in the meantime. The image appears in the chat, and your next message serves as a request for modification. This is the most important point: generation isn’t a random draw—it’s the start of a dialogue.

What makes a good image prompt

The difference between a usable visual and a generic image comes down to four pieces of information you either provide or omit.

Start with the subject: the object, scene, character, or situation to depict. Next comes the style, meaning the expected rendering: realistic photography, vector illustration, watercolor, minimalist and clean, flat design, or 3D rendering. The mood covers colors, lighting, and atmosphere: “Warm tones, golden sunset light,” “Blue and white palette, professional atmosphere,” “Black and white, high contrast.” Finally, composition sets the framing: “Top-down view,” “Centered portrait on neutral background,” “Wide-angle panoramic view,” “Close-up on hands.”

Omitting any of these four elements leaves the model to decide for you, resulting in generic output. “Make me a beautiful image” is the surest way to fail. Add the aspect ratio: ChatGPT Images produces images in any aspect ratio, which you specify in the prompt (16:9, 1:1, 9:16) or using the aspect ratio selector.

Four professional use cases

For a presentation, describe the concept to represent and specify the slide format.

“Create a minimalist illustration representing a company’s digital transformation. Flat design style, blue and orange colors, white background. 16:9 format for a slide.”

For a blog post, the logic is the same, but the image must work as a banner: “Generate a header image for an article about AI in business. Professional photography style, bluish tones, abstract elements suggesting data connections. Landscape format.”

For social media, include your brand guidelines: “Create a visual for a LinkedIn post announcing the launch of a new consulting service. Modern and professional style, navy blue and gold. Square format 1:1.” Aspect ratio is the variable people forget most often, and the one that renders an image unusable.

For a logo, describe the symbol rather than the general impression, and request a transparent background, which ChatGPT Images can produce: “Create a minimalist logo for a foodtech startup called ‘GreenBite.’ Symbol of a stylized leaf integrated into a fork. Green and white, transparent background.” To get started faster, the templates available under Images > Templates (poster, logo, etc.) ask the right questions for you.

Iterate rather than restart

The tool’s strength lies in iteration: you keep what works and correct the rest, message by message. “Replace the colors with navy blue and white,” “Keep the same composition but switch to vector illustration,” “Add the text ‘Innovation 2026’ at the bottom of the image.” For localized editing, select the image, outline the area with the selection tool, and describe the change; OpenAI warns that selection isn’t always precise and edits may bleed. This mechanism also applies to your own visuals: upload an image and describe the desired transformation.

Three recent features are worth knowing. You can request an animated looping GIF by describing what should move (feature rolling out). You can save a reference photo, such as your dog or product, under a name, then call it in another conversation using @ followed by that name. And on mobile, a finger-drawn sketch (@ then Sketch) can serve as a starting point.

All your images are stored in Images, where you can copy, save, or share them. To delete an image, you must delete the conversation that created it.

What the generator doesn’t guarantee

Proofread any text embedded in an image character by character before publishing. For a consistent series, repeat style details in every request or rely on a reference photo. Regarding rights, OpenAI’s European Terms of Use assign you ownership of generated content to the extent permitted by applicable law; they also clarify that another user may obtain a similar result. Images created by ChatGPT carry C2PA provenance metadata indicating their origin. Finally, requests violating OpenAI’s usage policies are refused.

Your turn

One visual, three versions. Generate a visual for a real business need, specifying subject, style, mood, composition, and aspect ratio. Request a color variant, then perform a localized edit using the selection tool. Finally, produce three versions of the same subject—in realistic photo, illustration, and minimalist styles—and compare what each conveys. If any version contains text, proofread it letter by letter.

Official sources: Images in ChatGPT (OpenAI Help Center: ChatGPT Images 2.6, plans, creation, editing, GIFs, references, templates, management) · What is ChatGPT: FAQ (C2PA metadata) · Europe Terms of Use (ownership of generated content) · ChatGPT Pricing (creation limits by plan)

Key takeaways

• ChatGPT Images creates and edits images directly in the chat across all plans
• A good image prompt specifies subject, style, mood, composition, and aspect ratio
• Iterate message by message, or edit a specific area using the selection tool
• GIFs, reference photos called with @, and mobile sketches complement creation
• Generated content belongs to you per OpenAI’s terms but isn’t necessarily unique; proofread all embedded text