Now live
GPT Image 2.5: Sharper Detail and Precise Edits, up to 4K
OpenAI released GPT Image 2.5 on September 8, 2026, and it is live on A2E today. Change only what you ask for, keep the subject from your reference photos, and export at 1K, 2K or 4K — transparent PNG included.
What GPT Image 2.5 puts out
Posters, product shots, interface mockups and editorial art — work where layout, reference and detail all have to hold. A sample of the work the model is built for.
Four things it does best
The first three are OpenAI’s own claims for this release; the fourth is what A2E exposes. What you get is whatever your own generation returns.
Change only what you ask for
OpenAI says GPT Image 2.5 is better at changing only what you asked for and leaving every other detail alone, even with complex subjects and backgrounds, and that it follows instructions more reliably across rounds of edits. In practice: one sentence, one change — recolor, move, add a line — without starting over.
Your subject, kept from the reference
OpenAI says the model preserves subjects from reference photos better. On A2E you can add up to 16 references for product shape, character identity, color, pose, composition or style — say which image controls what, and the product keeps its shape and color instead of being reinterpreted.
Layouts that hold together
OpenAI says it handles more complex layouts, including transparent backgrounds, with better style fidelity. For posters, packaging and interface mockups, put the exact wording in quotation marks and say where it belongs — then check every word before anything ships.
1K, 2K or 4K, square to 21:9
Choose 1K, 2K or 4K across 13 aspect ratios, from square and portrait to landscape and 21:9. Decide the channel first — a story, a banner, a print — then pick the ratio and resolution that suit it.
Pick the version first: Flare or Sunburst
GPT Image 2.5 comes in two versions, and A2E offers both.
Flare · everyday creation
OpenAI positions Flare as the default for most work: higher quality than GPT Image 2 and faster to return. Use it to explore directions and to run volume.
Sunburst · refined visuals
Sunburst spends longer on each generation for a higher level of precision, aimed at work that goes through several rounds of edits under tight control. Switch to it once the direction is settled.
Built for production work
Four kinds of work where placement, reference and color decide whether the image is usable.
Ads and social creatives
Headline, product hero, supporting copy and call to action arranged in a deliberate visual hierarchy — campaign concepts you can put in front of a client.
Product and e-commerce visuals
Place a referenced product in new scenes while keeping its color and recognizable details, for listings, banners and launch concepts.
UI mockups and structured graphics
Landing pages, dashboards, app screens, menus and infographics with defined sections and readable labels.
Posters, covers and editorial art
Typography, illustration, photography and multi-subject composition combined for posters, covers, explainers and branded content.
Four steps to your first image
This is the actual order of operations on A2E. Nothing extra to install or activate.
1. Describe the goal and layout
Write the subject, composition, style, colors, lighting and any exact text. Prompts can run to 20,000 characters, so a full creative brief fits.
2. Choose version, ratio, resolution and background
Flare for everyday creation, Sunburst for refined visuals; then set ratio, resolution and background. Decide where the image will be published first — that settles the ratio.
3. Add reference images
Optionally upload up to 16 product, character, composition or style references, and say what should be kept from each one.
4. Generate, review and refine
Check spelling, subject count, placement, brand colors and small details. When a result is close, correct the one thing that is off instead of rewriting the brief. Finished images land in My Results.
What you can set per generation
These are the controls GPT Image 2.5 exposes on A2E today — enough to judge whether it clears your delivery bar.
| Setting | What you get |
|---|---|
| Version | Flare for everyday creation, Sunburst for refined visuals. |
| Aspect ratio | 13 ratios, including Auto, 1:1, portrait, landscape and 21:9. |
| Resolution | 1K, 2K or 4K. Four ratios — 27:16, 16:27, 9:8 and 8:9 — are 1K only; every other ratio, Auto and 1:1 included, goes up to 4K. |
| Reference images | Up to 16 per generation. PNG, JPG, JPEG or WebP, 10 MB or less each. |
| Prompt | Up to 20,000 characters — room for a full brief, not just a sentence. |
| Background | Auto, Opaque or Transparent. |
| Output format | PNG with Auto and Transparent backgrounds; Opaque can also be saved as JPEG. |
| Access | Paid accounts only. |
AI output can still get a word, a count or a logo wrong. Check labels, claims and brand marks before anything ships.
Four habits that save a retry
Write a production brief
Use this order: goal, subject, composition, environment, style, lighting, color palette, exact text, output constraints. Split complex requirements into short sentences.
Quote exact copy
Put visible words in quotation marks, keep them short, and state placement and priority — for example, headline at the top, product centered, call to action at the bottom.
Assign every reference a job
Say which image controls identity, product shape, palette, pose, layout or style. Two references fighting over the same thing is the usual cause of a muddled result.
Change one variable at a time
When a result is close, change only the weak part of the prompt. Small, controlled revisions keep the composition and make it obvious what improved.
Docs and API
Building GPT Image 2.5 into your own product? Endpoints, parameters, authentication and billing are in the A2E developer docs, and MCP access is available there too.
Why run GPT Image 2.5 on A2E
Runs in the browser
No GPU, no install and no API key. GPT Image 2.5 draws on your A2E account like every other model on the site.
Every model, one account
GPT Image, Nano Banana, Seedream, FLUX and Z-Image sit behind one account and one balance. Run the same brief through several and keep the best.
From still to finished asset
Image editing, upscaling and image-to-video live next to generation, so a still can become a banner, a product page or a clip without leaving the site.
FAQ
What is GPT Image 2.5?
GPT Image 2.5 is OpenAI’s image generation and editing model, released on September 8, 2026 alongside ChatGPT Images 2.5. It comes in two versions, Flare and Sunburst, and A2E offers both. You write a brief, optionally add up to 16 reference images, and download the result at up to 4K.
Flare or Sunburst — which should I use?
Start with Flare: OpenAI positions it as the default, with higher quality than GPT Image 2 and faster results. Move to Sunburst when you need the extra precision for work that goes through several rounds of edits — it takes longer per image.
Which aspect ratios and resolutions can I choose?
There are 13 aspect ratios. 27:16, 16:27, 9:8 and 8:9 support 1K only; every other ratio, including Auto and 1:1, supports 1K, 2K and 4K.
How do I get accurate product colors?
Attach reference photos of the product and name the colors in the prompt, ideally with a concrete reference such as a brand color or a material. A color you state outright is far easier for the model to hold than one it has to infer from a description.
Is GPT Image 2.5 free to use?
You can start on A2E for free: new accounts get bonus credits on signup, no credit card is required, and you can claim extra credits every day by checking in. GPT Image 2.5 itself is available to paid accounts.
Can I use the images commercially?
Yes. Images generated on a paid A2E plan can be used commercially — ads, social media, client deliverables, e-commerce listings and print. You keep ownership of the output, with no watermark, no attribution requirement and no per-image royalties.
Where do my images go after generation?
Every finished job appears in the My Results tab beside the generator, where you can preview, download, share or delete it. The list keeps the last 3 days of generations, so download anything you want to keep.
Related AI image models on A2E
- GPT Image 2 and GPT Image 1.5 — the earlier generations, still available when you want to compare results on the same brief.
- Nano Banana Pro and Nano Banana 2 — fast generation and editing when turnaround matters most.
- Seedream 5.0 — ByteDance’s image model, worth comparing on the same brief.
- FLUX.2 and Z-Image — alternative image models worth running the same prompt through.
- Image to video — bring a finished still to life.