ChatGPT Images 2.5: What's New and How to Use It

ChatGPT's image generator has been renamed and rebuilt more than once: 4o image generation in 2025, then GPT Image models, and now ChatGPT Images, at version 2.5 at the time of writing. If you learned it a year ago, several habits are out of date. You no longer need to ask for "square, portrait or landscape" and hope; there is an aspect ratio picker. You no longer have to regenerate a whole image to change one area; there is a selection tool. This guide summarises what is current according to OpenAI's help centre and shows how to use it.

What changed

OpenAI also notes that image generation can take a few minutes for complex requests, and you can keep chatting while it works. Images can also be made and edited in Codex.

Where to find it

Ask for an image in any conversation, or select More and then Images to open the dedicated view, which holds Templates and your library. ChatGPT Images is on web, iOS and Android. GPTs with Image Generation enabled in their capabilities use the same new model.

Images with thinking

On Plus, Pro and Business plans, ChatGPT can think through a request before generating, which helps with prompts that have many constraints: a diagram with labelled parts, a poster with exact copy and layout, or an image grounded in facts from the conversation. OpenAI's developer documentation for the underlying GPT Image models lists the remaining limits honestly: complex prompts can take up to two minutes, precise text placement can still slip, recurring characters or brand elements can drift across generations, and exact layout control is not guaranteed. Thinking reduces those problems; it does not remove them.

Templates and sketches

1. Templates: open the sidebar, select Images, then Templates. Choose a category and template, customise the details and send. ChatGPT may ask follow-up questions, for example about visual style, and may offer an upload slot for a reference image. Templates are not yet available in Work mode.
2. Sketch (mobile app): in the message box type @ and select Sketch, draw, confirm with the tick, then describe the finished image, for example "Turn this sketch into a watercolour painting of a red house surrounded by trees."

Both are faster routes to a composition you can describe in shapes more easily than in words.

Editing with the selection tool

Select an image to open the editor. Use Select to highlight the area to change, then describe the change; or skip selection and describe the edit in the conversation, naming the area in your prompt. The editor also has an Aspect ratio option to regenerate the image in another shape, plus Undo, Redo, Cancel and Save. OpenAI notes that highlights are not always precise and edits may extend beyond the selected area. Full walkthrough in the ChatGPT image editor guide.

The Images library and sharing prompts

Everything you create is saved under Images on web and mobile, where you can copy, save or share each image. Deleting an image means deleting the conversation it was made in. On mobile you can share an image's prompt as a template link (Share, Prompt template, Copy link), so someone else can make their own version. For client work, still keep the prompt and file together outside ChatGPT, because a deleted conversation removes the image from the library too.

What it means for older prompts

Prompts written for earlier versions still work, but some workarounds are now unnecessary. You no longer need "square format" or "portrait orientation" phrases if you use the picker. You no longer need to describe the whole scene again to change one object; select it instead. And very long prompts benefit more from thinking than from extra adjectives. If a saved prompt suddenly gives different results, that is the model update, not your wording: rerun it a few times and adjust one detail at a time.

Prompting habits that still apply

• Describe the image like a brief: purpose, subject, setting, style, light, framing and any exact text in quotation marks.
• Set the aspect ratio with the picker rather than leaving it to chance.
• Edit in small steps and say what must stay the same.
• Ask for a transparent background when you need a cut-out.

The fundamentals are covered in the ChatGPT image generation guide, and text-heavy images in text and typography.

Try it

A clean infographic poster explaining how sourdough rises, in four labelled stages from left to right: "Mix", "Rest", "Fold", "Bake". Warm cream background, simple flat illustrations, dark brown sans-serif labels, no other text.

Generate it at a 3:2 ratio with the picker. If you are on a paid plan, run it again with thinking and compare how well the four labels and their order hold. Then use Select on one stage to change only its illustration.