Chat GPT Image: How to Create and Edit Images in ChatGPT (2026 Guide)

That last point matters, because many older guides on this topic were written for DALL-E 3, which OpenAI has since retired from ChatGPT, and some repeat fixed daily limits that no current OpenAI page states. Below is what OpenAI's own pages say today, how to get good results, where the feature falls short, and when a different tool makes more sense. Every fact here links to the page we read on the check date above.

What ChatGPT Images can do in 2026

ChatGPT Images is not a separate app. It sits inside ChatGPT on the web, iOS and Android, and OpenAI says you can also generate and edit images in Codex. You keep chatting while an image renders, which OpenAI warns can take a few minutes for complex requests.

According to the help center, the feature can:

• Create images from text. Describe the image, or open More and pick Images before typing.
• Edit uploaded photos. Upload an existing picture and describe the change.
• Add text and fine details. OpenAI lists adding text, adding details within the image and making the background transparent as supported instructions.
• Change the aspect ratio. Use the aspect ratio picker, or simply write the ratio in your prompt.
• Start from templates. Templates for things like posters and logos live under Images > Templates in the sidebar. They are not yet available in Work mode.
• Turn a sketch into an image. On mobile, type @, choose Sketch, draw, then add instructions.

Every image you make is saved automatically under Images, where you can copy, download or share it. One detail worth knowing: to delete an image from that library, you have to delete the conversation it came from.

Custom GPTs with the Image Generation capability turned on also use the new model, so a GPT someone built for you will not be stuck on the retired DALL-E version.

Images with thinking

There is a second mode called Images with thinking. OpenAI lists it for Plus, Pro and Business, with Enterprise and Edu "coming soon". The pricing page describes Plus as offering "More complex and accurate image creation", which lines up with this mode being the main reason to upgrade if images are your priority.

How to create a chat gpt image step by step

The basic flow is short. The skill is in the prompt, which we cover in the next section.

1. Open ChatGPT and start a new chat, or open Images from the sidebar.
2. Describe the picture in one to three sentences: what it is for, the subject, the setting and the style.
3. If you need a specific shape (a 9:16 story, a 16:9 banner), pick it in the aspect ratio picker or write it in the prompt.
4. Wait for the render. You can keep working in the chat meanwhile.
5. Ask for changes in plain language, or select the image to open the editor.

Editing a specific area

The editor gives you Select, Aspect ratio, Undo, Redo, Cancel and Save. Highlight the part you want changed, then describe the edit. On mobile, the image controls also include Resize and Remove.

OpenAI is unusually candid here: "Highlights are not always precise, and edits may extend beyond the area you selected." In practice that means you should compare the whole image after each edit, not just the area you painted. If a face, a logo or a line of text sits near your selection, say in the prompt that it must stay unchanged.

Prompt tips that OpenAI itself recommends

OpenAI's Academy guide on creating images (published 10 April 2026) gives more practical advice than most third-party listicles. The core idea is that a good prompt is clear rather than long. In most cases, one to three sentences are enough, as long as they cover the purpose of the image, the main subject, what is happening, where it takes place and the visual style.

A few rules from that guide are worth copying directly into your habits:

• Be concrete about light and materials. Academy's example: "soft natural light from a window on the left" works better than "beautiful lighting".
• State constraints out loud. If you do not want extra text, logos or visual changes, say so.
• Lock what should not change when editing. The suggested pattern is "Change only X. Keep everything else exactly the same."
• Treat text in images carefully. Put the words in quotes or ALL CAPS, specify font style, size, color and placement, keep it short, and spell unusual brand names letter by letter.
• For dense layouts, such as infographics with lots of labels, ask for "sharp text rendering" and plan to polish the result in a design tool.

Here is how those rules change a weak prompt:

Our own view: the third row is the riskiest. OpenAI itself tells you to consider polishing dense text in a design tool. That is a fair signal not to send a ChatGPT poster to print without proofreading every letter.

Free vs paid: what each ChatGPT plan includes for images

OpenAI's pricing page describes image allowances in words, not numbers. On 25 October 2026 it listed the following for individual plans. Prices are as displayed to a visitor from Indonesia and may differ in your region; check our ChatGPT pricing page for the current price and its check date.

What this table does not tell you is the actual number of chat gpt images per day on each tier. OpenAI does not publish it, and limits can shift with demand. If you see a fixed daily number for Free users elsewhere, look for the OpenAI page it comes from; we could not find one on the help center or the pricing page on the check date. We would rather tell you "not published" than repeat a number we cannot source.

Developers have a separate route: the GPT Image models are listed on OpenAI's API pricing page, and OpenAI's billing help says API usage is billed separately from your ChatGPT subscription. If you plan to generate images in bulk inside your own product, compare API pricing rather than chat plans. Our AI API overview explains how that billing usually works.

Labels, provenance and the rules you agree to

Every image you download from ChatGPT carries origin signals. OpenAI's provenance page says images generated with ChatGPT, Codex or the OpenAI API include both Content Credentials (the C2PA metadata standard) and SynthID watermarks.

Two practical consequences follow:

• Metadata is fragile, the watermark less so. OpenAI notes that Content Credentials can be lost through everyday actions such as taking a screenshot or converting a file. The invisible watermark is there to survive those changes where technically possible.
• A label is not a guarantee. OpenAI says provenance signals do not prove that content is accurate, unedited, legally owned or shown in the right context.

On use, OpenAI Academy says you are not required to credit OpenAI when you publish a generated image, though you may. It also asks you to be careful with likenesses: if you generate images of real people, including yourself, use a reference photo and make sure you have permission. And when in doubt, ask for a "generic" or "ownable" design instead of imitating a specific brand or artwork. All of this sits under OpenAI's usage policies, and the help center explains how to report images that may break them.

ChatGPT Images vs other AI image generators

ChatGPT is convenient because it lives where you already write. It is not the only free option, and for some jobs it is not the best fit. Here is how the entry points compare, using prices from each vendor's page as recorded in our directory on 25 October 2026.

How we would choose, based on what each vendor documents rather than on our own render tests:

• You already pay for ChatGPT and need quick visuals for a doc or a post: stay in ChatGPT. Having the image and the text in one thread is the real advantage.
• You want a free tool with a published daily number: Bing Image Creator is the only one in this table that states one.
• You want a Google ecosystem alternative at a similar entry price: see our ChatGPT vs Gemini comparison.
• You want an alternative built for stylized art: see our Midjourney alternatives, or browse the full AI image generation shortlist, and if you want motion next, the image to video tools.

A contrarian note: "best image generator" rankings age in weeks. Model names change, limits change, and a pricing page can rewrite its wording overnight. That is why every price on Newly Stack shows its source link and check date, and why we drop any fact older than 90 days from comparisons until it is re-checked.

Can ChatGPT Images handle Bahasa Indonesia?

Indonesian is on ChatGPT's official list of supported interface languages, so you can use the app in Indonesian. Whether the image model follows Indonesian prompts well, and whether it spells Indonesian words correctly inside an image, is a separate question that OpenAI does not document.

This is exactly the gap our Bahasa Indonesia benchmark is built to close. The NS-ID benchmark scores AI text models on real Indonesian tasks using a fixed test set, re-run every month; it does not cover image generation. In our separate test lab, OpenAI GPT Image 2 (through the API, not the ChatGPT app) passed all 5 short Indonesian photo-editing prompts on 25 October 2026, but we have not tested long Indonesian prompts or in-image Indonesian text yet. Until it is, our advice for Indonesian marketers is simple: keep in-image text short, follow OpenAI's own tip of spelling unusual words letter by letter, and proofread every label before publishing.

If you mainly need a general assistant rather than an image tool, our model leaderboard and the list of ChatGPT alternatives are better starting points. For no-cost options across categories, see free AI tools, and check current deals before paying full price for any plan.

Bottom line: chat gpt image creation in 2026 is a capable, convenient feature available on every ChatGPT plan, with editing, templates and provenance labels built in. Its weak spots are undocumented free limits, imprecise selection edits and dense in-image text. Start on Free, learn the prompt patterns above, and upgrade only once you hit the limit often enough to care.