How to Use GPT Image 2.5 (7 Steps)

TL;DR — GPT Image 2.5 is live inside ChatGPT across every tier, so you don't need to turn on any special setting. You can start with a simple prompt, pick templates, edit specific parts with the Select tool, or sketch ideas on your phone. It runs in the web browser, on mobile apps, and in Codex.

Here's how to create your first image in under ten minutes.

You've got the new model in front of you, but it isn't obvious which tools are worth your time. The interface has several ways to start, from plain chat prompts to full templates and sketch tools.

Here's the exact order to try them in, taken directly from OpenAI's own instructions. Follow along step by step to turn your ideas into clean pictures.

What you need before you start

Getting started takes very little setup:

• A ChatGPT account on any plan tier.
• The ChatGPT mobile app on iOS or Android if you want to draw sketches.
• A modern web browser if you're working on a desktop computer.
• Nothing to switch on, install, or buy before you begin.

Step 1: Open ChatGPT Images

You don't need to hunt through hidden settings, because that's all handled automatically. According to OpenAI's help centre article on images in ChatGPT, ChatGPT Images is available on all tiers. It works on the web, iOS, Android, and inside Codex.

Here's how to open it:

• Ask ChatGPT directly to create an image inside any regular conversation.
• Or open the sidebar, pick More, and tap Images.

Images with thinking is available on Plus, Pro, and Business tiers, and it's coming to Enterprise and Edu. Making a picture may take a few minutes if your request is complex, but you won't be locked out—you can keep chatting while it works.

The ChatGPT Images panel and its templates, captured 2026-09-09.

Step 2: Write a prompt that names the details that matter

The model responds best when you name the exact elements you care about most. Don't worry about complex prompt tricks; just state what you want to see.

Here's what ChatGPT Images can handle well when you name it clearly:

• Add specific text strings directly into your scene.
• Place fine details inside particular parts of the picture.
• Make the background transparent if you need a cutout file.

So name those exact details in your prompt box before you hit send.

Step 3: Pick your aspect ratio

Your aspect ratio is the relationship between the width and height of your image. Picking the right shape early saves you from redoing your work later.

You have two ways to set the shape:

• Choose your shape using the Aspect ratio picker inside the editor controls.
• Or tell ChatGPT the shape you want directly inside your prompt text.

You can stick to a standard square, or stretch the frame wider than it is tall, or taller than it is wide.

Step 4: Edit part of the picture without redoing it

You don't have to throw away a good generation just to fix a single mistake. You can use inpainting, which lets you pick part of an image and describe what should replace it.

Here are the two ways to edit:

• Tap the Select tool, highlight the area you want to change, and describe the fix in the chat box.
• Skip the tool completely and simply describe the edit and the specific area in the conversation panel.

Here's what the editor controls do when you're working on an image:

• Select: highlight the area you want to edit.
• Aspect ratio: regenerate the image in a different shape.
• Undo or Redo: revise your selections.
• Cancel: discard changes and start over.
• Save: keep your changes when finished.
• Mobile tools: use Resize and Remove, or add comments directly on an image.

Keep OpenAI's honest warning in mind: highlights are not always precise, and edits may extend beyond the area you selected.

Step 5: Start from a template instead of a blank box

If you don't want to start from a blank prompt, use pre-made layouts. Templates give you structured starting points for items like posters, merch, flyers, and product photos.

Here's the click path to open them:

• Open the sidebar on the left.
• Tap Images, then pick Templates.
• Browse the categories and choose a template like Poster or Merch.
• Add your specific details, such as the purpose of the poster and any text to include, then send.

ChatGPT may ask follow-up questions where you can pick suggested options or type your own. If an upload option appears, you'll be able to attach a reference image. Note that Templates aren't available in Work mode yet.

Step 6: Draw a sketch and let the model finish it

You can turn rough doodles into finished artwork using the built-in drawing tools.

Here's the mobile app flow:

• Type @ in the message box and pick Sketch.
• Draw your shape, picking colours, erasing, or undoing strokes as you go.
• Tap the checkmark to confirm your drawing.
• Describe what you want made from it, like asking to turn a house sketch into a watercolour painting of a red house surrounded by trees.

Here's the inconsistency: OpenAI's launch post for ChatGPT Images 2.5 says to type "@Sketch" in the box and links to a web page without calling it mobile-only. But the help centre describes Sketch strictly as a mobile app feature. We haven't tested it on desktop ourselves, so try typing @ in your message box on whatever device you're using.

Step 7: Save, share, or hand over the prompt

Every picture you generate is saved automatically under the Images section on both web and mobile.

Here's how to manage your finished files:

• Use Save to download the image file to your device.
• Use Copy to grab the image directly onto your clipboard.
• Use Share to send the image to another app.
• Share your prompt on mobile by tapping Share, picking Prompt template, and tapping Copy link so others can run the idea.

Here's the catch: to delete an image from My images, you must delete the conversation where you made it. On the web, hover over the conversation title, open the more options menu, and tap Delete. On iOS and Android, press and hold the conversation title, then tap Delete.

Settings that matter if you use the API

If you call the model through the API instead of ChatGPT, here are the settings you need to know:

• The model ids are gpt-image-2.5-flare for general use and gpt-image-2.5-sunburst for editing precision.
• You can set quality to low, medium, high, xhigh, max, or auto, as documented in OpenAI's image generation guide.
• Standard sizes are 1024x1024 square, 1536x1024 landscape, and 1024x1536 portrait.
• Custom sizes need multiples of 16, aspect ratios from 1:3 to 3:1, edges under 3840 pixels, and total pixels between 655,360 and 8,294,400 (sizes above 2560x1440 are experimental).
• Transparent cutouts need the background setting set to transparent with PNG or WebP output.
• Output defaults to PNG, but JPEG is faster, and both JPEG and WebP take compression settings from 0 to 100.
• Streaming costs an extra 100 image output tokens per partial image.
• Both models handle generation, edits, and inpainting, though some accounts must finish API Organization Verification first.

What GPT Image 2.5 still gets wrong

The model has clear limits you should plan around before you start:

• Complex prompts can take up to 2 minutes to generate.
• The model can still struggle with precise text placement and clarity.
• It may occasionally struggle to keep characters or brand elements consistent across several generations.
• It may have difficulty placing items accurately in structured layouts.
• Inpainting highlights aren't always exact, so edits can spill outside your chosen lines.

Every file created by the model carries C2PA metadata and SynthID invisible watermarking across ChatGPT, Codex, and API calls.

Run image models on Seadanse

If you want to try different generation styles in one browser generator, explore the tools on Seadanse.

To run GPT Image 2.5 on Seadanse, pick either GPT Image 2.5 Flare for fast everyday generations or GPT Image 2.5 Sunburst for precise editing at longer generation times. Type your prompt or add up to three reference images, choose 1K, 2K or 4K, and hit generate. They sit in the picker alongside the other image models:

• GPT Image 2 Generator for reliable OpenAI generations.
• Nano Banana 2 Generator for fast visual outputs.
• Seedream 5 Pro Generator along with Seedream 5 Lite for sharp detail.
• Grok Imagine Image 2 for alternative visual styles.

New accounts get free starter credits upon sign-up without needing a credit card. Check out Seadanse pricing if you want a credit pack without an ongoing subscription. If a generation ever fails or gets interrupted, your credits are refunded automatically.

The image model picker in the Seadanse generator. GPT Image 2.5 is two entries here, Flare and Sunburst — 2026-09-09.

GPT Image 2.5 FAQ

Do I need to turn on a special setting to use GPT Image 2.5?

No. The model is built into ChatGPT across all user tiers. You can use it by asking for an image in chat or opening the Images tab in the sidebar.

Does the Sketch feature work on a desktop computer?

OpenAI's help centre lists Sketch as a mobile app feature, while its launch post mentions typing "@Sketch" without limiting it to mobile. We haven't tested it on desktop, so type @ in your message box to see if it shows up on your screen.

What happens to my images after I generate them?

Your pictures save automatically in the Images section on web and mobile. If you want to delete an image, you must delete the entire chat conversation it was created in.

Summary

Here's the quick recap of the seven steps:

1. Open ChatGPT Images via chat or the sidebar menu.
2. Write a prompt that names your text, background, and specific details.
3. Pick your aspect ratio in the picker or directly in chat.
4. Edit specific areas using the Select inpainting tool or plain text instructions.
5. Pick a template for structured projects like posters or flyers.
6. Draw a sketch on mobile using @Sketch to guide the layout.
7. Save, copy, or share your final pictures and prompt links.

You can try it directly on the GPT Image 2.5 page alongside other leading image generators.

We read OpenAI's official help documentation, launch announcements, and API guides on 9 September 2026 to bring you these exact steps.

Related articles

• What Is GPT Image 2.5? — everything OpenAI shipped on 8 September.
• GPT Image 2.5 Flare vs Sunburst — which of the two API models to call.
• Is GPT Image 2.5 Free? — what each ChatGPT plan actually gets.