Poster
Prompt
A bold poster for a jazz night, the headline "Blue Hour Sessions" in big serif letters, a trumpet player in silhouette, deep blue and gold
GPT Image 2.6
Generate and edit images with GPT Image 2.6 Flare and Sunburst, the API models from the ChatGPT Images 2.6 release. Switch between fast drafts and detailed edits in one creator.
Flare & SunburstGenerate and editReference imagesHD to 4K
Enter a prompt to generate
Try an example
Flare vs Sunburst, Same Prompt
Each pair uses one prompt and the same settings: 3:4 at HD. Flare is on the left, Sunburst on the right.
Product
Prompt
A glass perfume bottle with a gold cap on a rough grey stone, soft window light from the left, a few water drops, clean product photograph, no text
Character
Prompt
A friendly red panda character in a knitted scarf holding a paper lantern, soft 3D render, warm evening light, plain cream background
Example images were made in this creator with the prompt shown. Results vary from run to run.
Flare or Sunburst?
Same price per image, same inputs. Pick by how many ideas you are testing and how careful the result must be.
Flare
Built for fast iteration
- Try many prompt ideas and compare them quickly
- OpenAI says Flare at low quality can return an image in as little as six seconds
- Runs at medium quality here, for cleaner drafts
- A good first pick while layout, colors and style are still open
Best for Brainstorms, thumbnails and social posts
Sunburst
Built for detail and consistent edits
- OpenAI says Sunburst gives higher quality at medium and high settings
- OpenAI says it stays more consistent across several edits
- Use it when you refine one image step by step
- Move your best Flare prompt here for the final version
Best for Final visuals, product shots and multi-step edits
What's New in GPT Image 2.6
What changed from version 2.5, based on OpenAI's release notes and API docs. Version 2.5 is still available here.
| Version 2.6 | Version 2.5 | |
|---|---|---|
| Models | Flare, Sunburst | Flare, Sunburst |
| Speed | OpenAI says images generate 50% faster than before | — |
| Price per token | Same as 2.5 | — |
| Output per image | About 3 times the output tokens of 2.5 at medium quality, as we measured | — |
| Flare credits per HD image | 15 | 5 |
How It Works
Three steps in the creator above. The credit price shows on the Generate button before you start.
1. Write a prompt or add references
Describe the subject, the style and any words that must appear. Attach references, at most 14, and give each one a job.
2. Pick Flare or Sunburst, then the size
Use the switch under the title or the model menu. Choose the aspect ratio, HD, 2K or 4K, and one, two or four images.
3. Generate, then edit
Check the result at its real size. Download it, or open it in an editing tool. To keep refining, add the result as a reference and run Sunburst.
Specs Side by Side
Both models accept references and offer three sizes. They cost the same. They differ in speed and detail.
| Flare | Sunburst | |
|---|---|---|
| Best for | Fast drafts and quick variations | Detail and consistent multi-step edits |
| Resolution | HD · 2K · 4K | HD · 2K · 4K |
| Reference images | Up to 14 | Up to 14 |
| Credits per image | 15 HD · 28 2K · 49 4K | 15 HD · 28 2K · 49 4K |
Keep working on the image you create
Generate another image by a new prompt. Edit the one you have when the subject works and one part needs attention.
FAQs
What is GPT Image 2.6?
GPT Image 2.6 is OpenAI's image model family, released on November 4, 2026. It comes as two API models, Flare and Sunburst. Both take a text prompt and optional images, and return a new image. You can generate and edit with either one in the creator above.
Is this the same as ChatGPT Images 2.6?
It comes from the same OpenAI release. That name is the image feature inside ChatGPT. This page uses the matching API models. Potto is an independent product, not part of ChatGPT, and some ChatGPT features are not in the API.
Should I use Flare or Sunburst?
Use Flare for fast drafts and many variations. Use Sunburst for detail and for edits that must stay consistent. Both cost the same per image. Draft on Flare, then move the best prompt to Sunburst.
Which model is faster?
OpenAI reports a speed gain over version 2.5 and quotes its quickest timing for Flare at the lowest quality setting. Here both run at medium quality, so expect longer waits. Sunburst usually takes longer than Flare.
Can I still use the 2.5 models?
Yes. The 2.5 Flare, Flare High and Sunburst models stay in the model menu. OpenAI has not retired them. Pick the version per request.
Can these models edit my own photos?
Both models edit photos you upload. Add up to 14 reference images. Say what stays and what changes. A reference is a guide, not an exact copy. Check shapes, labels and faces against the source.
How do I keep edits consistent with Sunburst?
Upload your last image again as the starting point. Name what must stay, such as the face, the label or the layout. Change one thing per pass. Sunburst is the better pick for this kind of step-by-step work. Compare each new version with the one before.
How should I write a prompt for these models?
Lead with the subject, then the style, then the layout. Quote any text that must appear. Say what to leave out, such as extra text or logos. Keep one idea per image. Add a reference when a shape or a color must match.
Which sizes and formats are available?
You get HD, 2K and 4K, ten aspect ratios from 1:1 to 21:9, and PNG or JPEG files. A square image is 1024 by 1024 at HD, 1920 by 1920 at 2K and 2880 by 2880 at 4K. OpenAI marks 4K as experimental, so review each 4K result.
Can I make more than one image per request?
Yes. Choose 1, 2 or 4 outputs in the settings. Each image costs the same, so four images cost four times one. Compare them side by side and keep the strongest.
How many credits does each image cost?
Credits depend on the size and the number of images. The Generate button shows the total before you confirm. A failed request does not use credits. Both models cost the same.
Why do the 2.6 models cost more than 2.5?
The token price is the same, but each new image uses more of them. At medium quality, we measured roughly three times as many tokens per image as version 2.5. Credits follow OpenAI's official price per image, so the newer models cost more. Failed requests cost nothing.
Do I need an account to start?
You can type your idea, add references and pick a model before you sign in. Generate opens the sign-in dialog. Everything you entered stays in the creator.
What if a request fails or is blocked?
A failed request releases its credits. The task explains why it stopped. Content blocked by OpenAI's safety checks is not retried automatically. Change the prompt, then try again.
Do the models get text right?
Short headlines often come out right, but OpenAI still lists text rendering as a limit. Read every word in the result. Keep long copy in your own layout. Put the exact words in quotes in the prompt.
Try GPT Image 2.6 on Potto
Choose a model, write a prompt and see the credit price before you generate.