Two doors lead to an image in ChatGPT: describe the picture in the conversation you are already in, or select More and then Images. Both reach the same generator, and both are open to every account.
ChatGPT Images runs on all tiers, on web, iOS and Android, and Codex generates and edits images too. One restriction: images with thinking is available on Plus, Pro and Business, with Enterprise and Edu coming soon. Everything below comes from OpenAI’s help pages; where a figure is missing, this page says so.
Where image generation lives
GPT Image 2.5 is documented for all ChatGPT, ChatGPT Work and Codex users across desktop, mobile and web. Inside the app there are two starting points, and a third outside it.
| Way in | How you open it | What you land on |
|---|---|---|
| Conversation | Ask in any chat | The picture inline, ready to edit |
| Images | Select More, then Images | The same generator, plus saved images |
| Codex | Generate or edit in a Codex session | The same, without leaving your editor |
Generation is on all tiers; images with thinking is not, because the help centre lists it for Plus, Pro and Business only. The platform list adds Codex to web, iOS and Android. The GPT Image 2.5 overview covers the model itself.
From prompt to finished picture
Write the prompt, set the ratio, and send. OpenAI gives generation one time estimate — an image “may take a few minutes, depending on the complexity of your request” — and notes you can keep using ChatGPT while it renders.
The same page lists three things the model will follow as instructions: add text to the picture, add details within the image, and make the background transparent.
An existing picture can be uploaded and then described, which turns the prompt box into an editing surface. Aspect ratio is not fixed: use the aspect ratio picker, or put the ratio in the prompt — the version that survives in a prompt you save.
Editing without starting over
Two ways are documented, and neither is ranked above the other:
- Select part of the image with the selection tool, then describe the changes in chat.
- Skip the selection tool and describe the edit directly in the conversation panel.
The first route is for precision; the second is faster when the change covers most of the frame.
The selection tool and the editor
The editor, in the help centre’s wording, “lets you add, remove, and update parts of your image.” The surface also carries regeneration at a chosen aspect ratio, undo and redo, and saving.
What a selection cannot promise
One limitation is stated plainly: “Highlights are not always precise, and edits may extend beyond the area you selected.”
That is a checking instruction as much as a warning. After each selected edit, look along the rim of the region you drew. If something outside the selection moved, delete that turn and resend with a tighter selection.
Sketch, Templates and Comments
Three named tools, three jobs:
- Sketch — on the mobile app, type @ in the message box, choose Sketch, draw the outline, confirm it, then add written instructions. The sketch carries layout, not detail.
- Templates — open the sidebar, go to Images and then Templates, pick a template such as a poster or a logo, then add your details. OpenAI’s note is explicit: templates are not yet available in Work mode.
- Comments — place a comment directly on an image to focus the next edit more tightly than a whole-frame instruction.
A quieter fourth change is prompt sharing: the prompt you used can be shared to run against someone else’s photo.
Two worked cases: create and refine
Both cases are copy-and-run: the first makes a picture, the second changes one region of an existing one.
Case one: a poster from a prompt
A poster is a fair test: ratio, empty space and lettering all have to be stated.
Format: A2 poster, portrait, 3:4.
Headline text, set exactly: "HARVEST MARKET" — large, centred, upper third.
Subhead text, set exactly: "Saturdays, 8am to 1pm" — small, centred, below the headline.
Subject: a shallow crate of late-summer vegetables on plain linen, eye level.
Light: soft warm morning light from the upper left, gentle falloff.
Empty space: keep the lower third calm and unbroken for a date stamp.
Style: screen-printed poster, two flat inks, visible paper grain.
Exclusions: no logo, no watermark, no second crate, no extra text.If the lettering returns wrong, change only the quoted line and resend; a prompt that rewrites the layout too teaches you nothing.
Case two: a background swap by selection
Here the subject is right and only the surroundings are wrong, so the edit is worth selecting.
Reference: the uploaded photo of one subject standing outdoors.
Selection: the whole background behind the subject, edge to edge and top to bottom.
Change: replace it with a seamless pale grey studio sweep and soft even light.
Keep: the subject, its outline, its proportions, its pose and the light already on it.
Do not: add props, text, reflections or a second person.
After: check the subject's rim for stray grey or clipped edges.That is the acceptance check the warning implies: look along the subject’s rim, where a swap leaves evidence.
What the help centre leaves out
The gaps matter as much as the documentation, because other pages fill them with numbers.
- Per-tier quotas. How many images a plan allows per day or month is not published, free or paid.
- Watermarks. There is no watermark statement at all, in either direction.
- Tool differences beyond thinking. Nothing is published on how tools vary by tier beyond the thinking split.
- Generation time. The only published timing is “may take a few minutes,” with complexity named as the variable.
A specific daily image count on any page is someone’s account of their own usage, not a vendor specification. Missing numbers stay missing here.
ChatGPT image FAQ
- Where do I generate an image in ChatGPT? Two places inside the app: ask in a conversation, or select More and then Images. Codex is a third route.
- Must I start a new image to fix one detail? No. Select the region and describe the change, or describe the edit in the panel.
- Does this work on a free account? OpenAI says ChatGPT Images is available on all tiers; images with thinking is limited to Plus, Pro and Business.
- Why did something outside my selection change? Highlights are not always precise, and edits may extend beyond the selected area. Reselect more tightly.
Where to go next
Four pages go deeper:
- Sketch covers drawing the layout before prompting.
- Templates covers preset formats, including the Work mode gap.
- Combine images covers prompts built from several uploads.
- Comment editing covers edits placed on the image itself.