GPT Image 2.5

An idea in your head. An image you can use.

Try OpenAI’s GPT Image 2.5 on Playground. Start with Flare for everyday images, or choose Sunburst for a more exacting brief. Type a prompt or upload a photo to edit.

Up to 3 free images every 30 days, shared across image tools. No sign-up. 1024 × 1024 PNG at medium quality.

How GPT Image 2.5 works

1. Start with a T-shirt idea

Describe the shirt, the artwork, and the words you want on it. Here, a cream tee gets a coral sun and green “WEEKEND CLUB” lettering. Ask for a clean product photo to see the idea as a piece of merch, or upload your own design as a reference.
Start with a T-shirt idea

2. See your design on a product

Start with Flare to explore the concept, or choose Sunburst for a brief with more details to preserve. Check the lettering, graphic placement, and fabric texture. Try the same prompt with both to see which result suits your brand.
See your design on a product

3. Try another colorway

Save the PNG or continue in Playground’s editor. Ask: “Make the shirt butter yellow. Keep the print the same.” Compare the versions for your collection or shop concept, and check that the lettering and graphic stayed intact. Editor access and usage follow your account’s plan.
Try another colorway

Try your first T-shirt design

Have an idea for a small merch drop? Start with the shirt, the graphic, and the words you want on it. This prompt brings them together in a product image you can use to explore the direction. Check the lettering and print placement, then try a different colorway.

Design a cream T-shirt with a coral sun and green “WEEKEND CLUB” lettering. Show it as a clean product photo on a peach background.
Try this prompt free

The images above illustrate the workflow. Run the prompt to compare your own results.

What makes GPT Image 2.5 worth trying?

A good image model has to do more than make something attractive. It needs to follow the brief: put the headline where you asked, keep the product recognizable, and change the background without reinventing the subject. GPT Image 2.5 gives you two ways to approach that work. Flare is built for fast everyday generation. Sunburst puts the emphasis on capability and editing precision. Both work with text prompts and image references.

Read OpenAI’s model notes for Sunburst and Flare.

Turn a brief into a first draft

Describe the audience, subject, composition, and mood. Start with a social post, a campaign concept, or a product image, then judge how closely the result follows your idea.

Make a specific change to a photo

Upload an image and say what should change and what should stay. A request such as “replace the table with pale marble; keep the bottle and its label” is a useful editing test.

Compare two approaches to the same job

Try one prompt with Flare and Sunburst. Look at the details that matter to you: the wording, the arrangement of objects, and whether a reference still looks like itself.

Sunburst vs Flare: where to start

Choose based on the job in front of you. Flare suits a quick first draft; Sunburst is worth trying when you need a more precise edit. Both can generate new images and work from an uploaded reference.

GPT Image 2.5 Sunburst and Flare compared
Your prioritySunburstFlare
Model focusCapability and editing precisionSpeed and everyday generation
A useful first testA product edit with several details to preserveA social image or a new campaign concept
Free Playground output1024 × 1024 PNG, medium quality1024 × 1024 PNG, medium quality
Try itTry Sunburst freeTry Flare free

Model positioning: OpenAI Sunburst documentation and OpenAI Flare documentation. The suggested tests are our practical starting points. Generation time and results vary with the brief.

Independent model rankings

GPT Image 2.5 on Arena

Sunburst and Flare hold the top two positions in these Arena snapshots. Here are the scores, along with the original leaderboards so you can check the evidence yourself.

Text-to-image generation

#1 · Sunburst
1,421 ± 13
#2 · Flare
1,399 ± 13
See Arena’s text-to-image generation leaderboard →

Single-image editing

#1 · Sunburst
1,520 ± 9
#2 · Flare
1,491 ± 9
See Arena’s single-image editing leaderboard →

Arena snapshot dated , checked . Both models are marked preliminary. The ± values are Arena’s reported uncertainty. Scores reflect human preferences, not accuracy percentages; rankings may change as votes accumulate. These scores evaluate the models, not Playground’s free trial settings.

Find the model that fits your next project

Use the same brief across a few models. Compare the details that matter for your project, from lettering to product photos.

OpenAI

GPT Image 2

An earlier GPT Image model for typography, text rendering, and instruction following.

Speed
~15-25s
Output
Up to 1024 × 1536
Pricing
Free — 3 / month

Best for

Posters, ads, packaging, infographics, and any image with legible text.

  • Reads dense prompts and complex layouts
  • Renders crisp typography and on-image text
  • Edits with strict prompt fidelity

Google Gemini 2.5 Flash Image

Nano Banana

The everyday workhorse. Lightning-fast iteration with strong reference fusion and character consistency.

Speed
~3-6s
Output
10 native ARs up to 1K
Pricing
Free — 3 / month

Best for

Rapid retouches, social posts, exploration, and blending multiple references.

  • Lightning-fast iteration loop
  • Strong reference image fusion
  • Great for character / style consistency

Google Gemini 3 Pro Image

Nano Banana Pro

The highest-fidelity Gemini model. 4K native output, sharper detail, and improved photorealism.

Speed
~25-40s
Output
Up to 4K, 10 ARs
Pricing
1 free / 30 days, no sign-up

Best for

Photoreal hero shots, large-format prints, and detailed scene composition.

  • 4K native resolution output
  • Best-in-class photorealism
  • Improved text + spatial reasoning

Try Muse Image — our free-user Auto default, at 1 credit each

See the LMArena text-to-image leaderboard

Frequently Asked Questions

Last updated: