ChatGPT Images 2.5: How to Create Images and What Changed for Brands

To create an image in ChatGPT, type what you want into the chat, or open the "Images" section in the sidebar and enter a prompt. According to OpenAI's help page, ChatGPT Images is available on all plans, on web, iOS and Android, so you can generate images with a free account too. The latest model is ChatGPT Images 2.5, announced on 8 September 2026; it brings more accurate editing, results that stay closer to a reference photo, and faster generation.

How do you create an image in ChatGPT?

The quickest way is to ask directly in the chat, for example: "Create a simple 1:1 product image for our coffee brand on a light wooden table." Depending on how complex the request is, generation can take a few minutes, and you can keep chatting in the meantime.

If you want a more organized workflow, you can use these options:

• The Images section: Open Images in the sidebar and start a new image with "Create Image". Every image you generate is saved there automatically.
• Templates: The Templates area in the Images section has categories such as posters, logos and product photos. When you pick a template and fill in the details, ChatGPT can ask follow-up questions for missing information such as style. OpenAI notes that templates are not yet available in Work mode.
• Sketch: In the mobile app, type @ in the message box and choose Sketch to draw a rough sketch, then turn it into an image together with a written instruction.
• Aspect ratio: Use the ratio selector or state the ratio in your prompt; ask for separate ratios for an Instagram post, a story and a banner.

OpenAI's older DALL·E GPT has been removed from ChatGPT; image generation and editing now happen only through ChatGPT Images.

How do you edit a generated image?

There are two methods. The first is to write in the chat what you want to change, for example: "Make the background marble, do not touch the product itself." The second is to click the image, mark an area with the selection tool in the editor and give an instruction for that area only.

OpenAI states plainly that selections are not always precise and that edits can spill outside the marked area. So name the elements you do not want changed in your prompt as well. On mobile, opening the image full screen also shows options to resize and remove objects. You can edit a photo you upload yourself in the same way.

What changed with ChatGPT Images 2.5?

According to OpenAI's announcement of 8 September 2026, the main changes are:

• People, products or places from a reference photo stay more recognizable in new scenes, with more natural lighting and texture
• Only the requested element is changed and the rest is preserved; earlier edits are not undone by later ones in a series of edits
• Better generation of complex layouts and transparent backgrounds
• Up to 50 percent lower generation latency compared with Images 2.0
• New features such as Sketch, templates, leaving comments on an image and sharing prompts

The new model has been rolled out to ChatGPT, ChatGPT Work and Codex users on all plans. For developers, there are two models in the API: GPT-Image-2.5 Flare, the recommended default for most uses, and GPT-Image-2.5 Sunburst, which offers finer control with a longer generation time. GPT-Image-1.5, which appears in older articles, is no longer the current model.

Is there a limit on image generation in ChatGPT?

OpenAI's image help page does not publish a fixed number of images per plan. The distinction it does make is this: ChatGPT Images is on all plans, while image generation with thinking (Images with thinking) is offered on the Plus, Pro and Business plans, with Enterprise and Edu described as coming soon. If your team generates images heavily, looking at business plans or the API is more predictable than relying on a single account.

How do you write a prompt for on-brand images?

A generic prompt produces a generic image. For images you will use for your brand, structure the prompt like a short design brief:

1. Placement and ratio: For example, "Instagram post, 4:5".
2. Subject: The product, the scene, and say so if there should be no people.
3. Visual language: Lighting, camera angle and style; give your brand colors as color codes.
4. Text in the image: Put the exact text in quotation marks, word for word. Always check non-English characters, such as Turkish letters, carefully.
5. What must not change: List the elements that must be preserved, such as the logo, packaging and product color.
6. Reference: If you have one, upload a real product photo and say "keep the product's shape and label exactly as they are".

Keep the prompts that work in a shared document. Working from templates makes images produced by different people look as if they came from the same brand. If your corporate color and typography rules are not written down yet, settling those first is more productive.

What should brands watch out for?

Ownership: OpenAI's terms of use assign ownership of the output to the user, to the extent permitted by applicable law. The same terms also state that output may not be unique and that similar results may be given to other users. So do not treat a logo generated in ChatGPT as a safe starting point for trademark registration on its own.

Provenance: OpenAI says it uses C2PA metadata and an invisible watermark to help identify images made with its tools. It can be detected later that an image was generated with AI.

Advertising rules: For campaigns in Turkey, an amendment to the advertising regulation that took effect on 1 August 2026 requires ads to state clearly when AI has been used in a way that significantly affects the consumer's decision, or when AI characters indistinguishable from humans appear. Showing an AI-generated copy of a real person as if they use or recommend the product is prohibited. If you advertise elsewhere, check the local rules.

Product accuracy: In e-commerce, an image that shows a product differently from reality comes back as returns and complaints. Change the background and scene with AI, but take the product itself from a real photo.

Before you publish: a short checklist

• Are the text and special characters in the image correct?
• Do the product's shape, color and label match the real thing?
• Are the brand colors and logo preserved?
• If a recognizable real person appears, do you have their permission?
• If the image will be used in an ad, is the use of AI disclosed where required?
• Have the prompt and source files been archived?

We cover how to fit AI into your content production in our post on generative AI content creation. If you would like to plan campaign and social media visuals together with a design team, take a look at our graphic design service.