ChatGPT Image Generator: How to Use Images 2.6 on Any Plan

In brief

The ChatGPT image generator is ChatGPT Images 2.6, released on 4 November 2026 and open on every plan, Free included. You describe a picture in any chat or open Images in the sidebar. OpenAI prints no image count for any consumer plan, and only Plus, Pro and Business get the thinking mode.

How do you make an image in ChatGPT?

Type what you want in any chat, or select Images and enter a prompt. OpenAI's help page says ChatGPT Images is available on all tiers, on web, iOS and Android. Five steps cover the whole job.

A silver OpenAI knot on a black plinth beside three blank photo prints in square, tall and wide shapes, with a small terracotta paintbrush resting on an easel on a second plinth.
  1. Open a chat or the Images page. On chatgpt.com or in the app, start a normal chat. Or select More and then Images.
  2. Describe the picture. Name the subject, the setting and the light. OpenAI says Images can follow instructions to "add text, add details within the image, or make the background transparent."
  3. Pick the shape. Use the aspect ratio picker, or write the ratio into the prompt.
  4. Wait, then refine. OpenAI says generation "may take a few minutes, depending on the complexity of your request," and you can keep using ChatGPT while it runs. Ask for one change per message.
  5. Save it. Open the image and select Save. Every image also lands in Images, and deleting the conversation removes it from there.

Three prompts that fit the way the tool reads text. Each names a subject, a shape and what must stay out.

  • A square product photo of a ceramic mug of black coffee on a walnut desk, soft window light from the left, shallow depth of field, no text.
  • A 3:2 poster for a neighbourhood book swap. The headline reads "Swap Saturday" in thick hand-cut letters, a stack of paperbacks below, warm orange and teal.
  • A flat sticker of a smiling cactus in a red pot with a white outline. Make the background transparent.

Keep the wording in quote marks short, as in the poster prompt. OpenAI's own API guide lists text placement as a known weak spot, so read every letter before you post the result.

Which model runs the ChatGPT image generator now?

ChatGPT Images 2.6 runs it, released on 4 November 2026. OpenAI's announcement says it makes images 50% faster than before, and its help page says it keeps details more consistent from one edit to the next. It added two things you can use today: looping GIFs and saved reference photos. DALL-E is gone, since OpenAI retired the official DALL-E GPT from ChatGPT on 30 August 2026.

ChatGPT's image model has been replaced four times since March 2025
2026ModelGPT-4o image generationChatGPT Images relaunchImages 2.0Images 2.5Images 2.6RetiredDALL-E GPT retired

The dates come from OpenAI's release notes and launch posts. GPT-4o image generation arrived on 25 March 2025. A new ChatGPT Images, which Wikipedia names GPT Image 1.5, followed on 16 December 2025 for Free, Go, Plus, Edu and Pro, with Business and Enterprise listed as coming. Images 2.0 landed on 21 April 2026, the first with a thinking mode, and TechCrunch reviewed it the same day. Images 2.5 shipped on 8 September 2026 with templates, sketch input and comment-based edits. Then came 2.6. That's one new model every four to five months. So when a guide you find still talks about DALL-E or GPT-4o, it's out of date, and you shouldn't trust its menus or its limits either.

Developers get two named versions, GPT-Image-2.6 Flare for fast drafts and GPT-Image-2.6 Sunburst for more detail. OpenAI says Flare can finish a low-quality image in as little as six seconds. ChatGPT picks for you: DataCamp reports that you can't choose between Flare and Sunburst in the app. OpenAI also says people now create 4 billion images a week across ChatGPT, Codex and the API.

How do you edit an image you already made?

Select the image and describe the change, or paint over the part you want changed first. OpenAI's help page gives two routes: use the selection tool and then describe the change in chat, or skip the tool and describe the edit directly. If you skip the tool, name the area in the prompt.

  1. Open the editor. Select an image in the conversation or in Images.
  2. Select an area. Choose Select and highlight the part to change. Undo and Redo fix a slip, and Cancel starts over.
  3. Describe the change. One change per message works best. OpenAI warns that "Highlights are not always precise, and edits may extend beyond the area you selected."
  4. Change the shape if you need to. Aspect ratio in the editor regenerates the image in a different ratio.
  5. Save. Select Save when you're done.

Four other tools came in with Images 2.5 and 2.6. Saved reference photos let you upload a picture, give it a name and call it later with @ and the name, as in "Make a watercolor portrait of @Pepper". Templates, under Images and then Templates, start a poster or logo from a category. On the mobile app you can type @ and pick Sketch, draw a rough shape and add instructions. And GIFs are rolling out: ask for a looping GIF, or upload a still and ask ChatGPT to animate it. ChatGPT can edit a GIF it made, but it can't edit an animated GIF you upload.

Uploads have their own cap. OpenAI's file upload FAQ says free users get 3 file uploads a day and other plans 80 every 3 hours, with a limit of 20 MB per image. Photo-specific steps, faces and what the selection tool misses are in can ChatGPT edit photos.

What does each ChatGPT plan get for images?

Every plan can generate images, and only the paid plans above Go get the thinking mode. OpenAI's pricing page words the allowance in adjectives, not numbers. The table is from that page and the help page, read on 15 November 2026.

PlanImage wording on OpenAI's pricing pageImages with thinkingPublished image count
FreeLimited image creationNoNone
GoMore image creationNoNone
PlusMore complex and accurate image creationYesNone
ProUnlimited and faster image creationYesNone
BusinessIncluded (the help page says all tiers)YesNone on the individual pricing page

Images with thinking is the paid extra. OpenAI's release notes say that, given more time to think, ChatGPT can plan and refine an image before it generates it, and that you get this by selecting a Thinking or Pro model. The help page lists it for Plus, Pro and Business and says Enterprise and Edu are coming. It names no limit for it either.

The words on that pricing page don't tell you when you'll be cut off. The countdown messages, the 720-hour reports on paid plans and what edits cost are all in ChatGPT image generation limit. No plan prints a number, and a footnote on OpenAI's pricing page says Pro's "unlimited" is subject to abuse guardrails, which is why you may still see a countdown on any of them.

What size are ChatGPT images, and can you choose one?

You can choose the shape in ChatGPT but not a pixel size, because OpenAI's help page names no resolution for ChatGPT output. It says you can generate images in any aspect ratio, using the picker or the prompt. As of 15 November 2026 the exact sizes are published for the API only, and the table sets the two side by side.

In ChatGPTIn the API (GPT-Image-2.6 Flare and Sunburst)
Aspect ratioAny, via picker or promptCustom sizes from 1:3 to 3:1
Recommended sizesNone published1024x1024, 1536x1024, 1024x1536
Largest edgeNone published3840 pixels; above 2560x1440 is experimental
Pixel budgetNone published655,360 to 8,294,400 pixels
File typesNot publishedpng, jpeg, webp; svg and gif on request
Transparent backgroundAsk in the promptbackground transparent with png or webp

The API rules come from OpenAI's image generation guide: width and height must be multiples of 16, and quality runs from low to max. GIFs take 2 to 16 frames at 20 to 2,000 milliseconds each. The same guide warns that complex prompts can take up to 2 minutes. If you need a hero image of an exact size, you have two routes that work: ask ChatGPT for the right ratio and resize the file afterward, or skip the app and call the API yourself, where the size is a setting. Billing there is separate from any ChatGPT plan, and DataCamp reports the output rate at $30 per million tokens, unchanged from Images 2.5.

What will the ChatGPT image generator refuse?

It refuses anything its policy checks flag before or after drawing, and it applies extra limits to real people, minors, sexual content and GIFs. OpenAI's Images 2.6 system card, published 4 November 2026, describes two layers. Upstream, "Requests deemed to violate policy are refused at this stage." Downstream, a safety model checks the inputs and the finished image, and blocks either one if it finds a violation.

The categories OpenAI tested are violence, sexual content, self-harm and political content, plus a long-tail group it names as religious figures, religious satire and "fabricated or compromising depictions of real people". Its March 2025 launch post is the plainest on people: it blocks requests for child sexual abuse material and sexual deepfakes, and when images of real people are in context it has "heightened restrictions," with "particularly robust safeguards around nudity and graphic violence." That post predates Images 2.6, so treat the details as a floor, not the current rulebook.

GIFs get stricter treatment. The system card says GIF-specific rules "block some content categories more broadly than in still images to account for inferred motion," and adds that safeguards may not catch every motion violation. Teen accounts see a reminder before some uploads, and a parent can control whether a linked teen creates or edits images.

Two things travel with every picture. OpenAI keeps C2PA metadata on Images 2.6 outputs and adds an invisible SynthID watermark, and its public verification tool at openai.com/research/verify can check supported files. If a request is refused, change the idea. Pages that promise a no-restrictions version of the ChatGPT image generator are selling a different tool.

How does it compare with Gemini and Claude?

ChatGPT and Gemini both generate and edit images on free accounts, and Claude generates none. The table is compiled from each vendor's own help pages, read on 15 November 2026, for individual plans.

ChatGPTGeminiClaude
Image modelImages 2.6Nano Banana 2, Nano Banana Pro redo on paid plansNone
On the free planYes, "limited"Yes, Nano Banana 2 is listed on every planNo generator
Edit an uploaded photoYes, with a selection toolYes, upload an image and ask for editsReads images, cannot change them
Download sizeNot published2K with a Google AI plan, 1K withoutNot applicable
Animated outputGIFs, rolling outNot covered on the help pageNone
Published daily countNoneNone on the help pageNone

Pick ChatGPT when the picture has words in it, when you want to refine one image over many turns, or when you need a GIF. Pick Gemini when you want a published download size (2K on a Google AI plan) or a second take on the same prompt. Claude isn't an option for pictures: Anthropic's vision guide describes image input only, and can Claude generate images lists the workarounds. Best AI image generators ranks the models by blind vote and price, and Gemini image generation limit covers Google's side.

What if you want images and Claude, GPT and Gemini in one place?

Whizi, which publishes this page, runs Claude, GPT and Gemini in one app, and it is not a copy of ChatGPT Images. Pro carries all three families, and its first week is $1.99. The free plan gives 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5, with no card.

Pictures are a separate allowance. The creative mode in the chat header switches the model picker to Whizi Image, Nano Banana, Flux and Stable Diffusion. Pro includes 100 images a month and the free plan 2. Nano Banana edits your own photos and takes up to 4 source images in one edit. How image generation works in Whizi lists the rest.

Whizi loses on two counts. It doesn't run OpenAI's image model, so there's no Images 2.6 and no thinking mode. The monthly count is small next to Pro's "unlimited" wording from OpenAI. It wins when your week mixes writing, a Claude answer and the odd image, and you'd rather pay for one login than three.

Workflow checklist
  • Start in a normal chat or in Images; there is no switch to turn on and no plan gate for plain image generation
  • Put the ratio in the prompt or use the picker, and resize afterward if you need exact pixels
  • Ask for one change per message and use Select for local edits, remembering that edits may spill past the highlight
  • Keep any words in the picture short, in quote marks, and read every letter before posting
  • Save images you want to keep: deleting a conversation removes its pictures from Images
  • Expect refusals around real people, minors and sexual content, and expect every output to carry C2PA and SynthID marks
Common questions

Frequently asked questions

Is the ChatGPT image generator free?

Yes, with limits. OpenAI says ChatGPT Images is available on all tiers, and its pricing page lists Free as "Limited image creation" with no number. Free accounts also get 3 file uploads a day, which limits how many photos you can upload to edit.

What model does ChatGPT use to generate images?

ChatGPT Images 2.6, released on 4 November 2026. It replaced Images 2.5 from 8 September 2026. OpenAI retired the official DALL-E GPT on 30 August 2026, so DALL-E is no longer an option inside ChatGPT.

Can ChatGPT make a GIF?

Yes, where it has rolled out to your account. Ask for a looping GIF and say what should move, or upload a still and ask it to animate. ChatGPT can edit GIFs it created, but it can't edit an animated GIF you upload.

Can ChatGPT edit my own photos?

Yes. Upload the photo, then describe the change or highlight the area with the selection tool. Edits can spill past the highlight. Steps and limits are in the guide on whether ChatGPT can edit photos.

Is there a ChatGPT image generator with no restrictions?

Not from OpenAI. Its policy checks run before and after drawing on every plan. Sites that advertise a no-restrictions ChatGPT image generator are separate tools, not OpenAI products.

Why is the ChatGPT image generator not working?

Usually a limit or a refusal. A limit message prints its own countdown and the fix is to wait or use another tool, as laid out in the image limit guide. A refusal means the request tripped a policy check, so change the idea. If nothing appears at all, check status.openai.com first.