GPT Image 2.5: Text That Reads, Edits That Stick, 4K That Ships

Fewer re-rolls than any image model we run: headlines spelled as written, the clock set to the time you asked, the background swapped without touching the subject. Flare is twice as fast as GPT Image 2 and sharper; Sunburst is the cleanest, most controllable output OpenAI has released — both at up to 4K, thirteen ratios, and transparent PNG on demand.

Fast default — better than GPT Image 2 at about half the latency. 1K–4K, transparent backgrounds.

0/2000

0 credits available

Sample image generated with GPT Image 2.5Sample — GPT Image 2.5 Sunburst

Your image appears here

Hit Generate to render with GPT Image 2.5. The result shows here as soon as it is ready, with a download link.

GPT Image 2.5 Is Two Models, and Picking the Right One Is the Whole Trick

OpenAI ships GPT Image 2.5 as a pair. Flare is the fast, lighter model — sharper than GPT Image 2 at roughly half the latency, built for social content, product shots, thumbnails, and anything you generate in volume. Sunburst is the heavier model tuned for quality and for precise, repeatable edits — the one you reach for when the image is going into a campaign, a listing, or print. Both render 1K, 2K, and 4K, follow instructions literally (the clock reads 5:15, the glass is half-full), render legible text, and can return a transparent background. Both take a text brief, or a text brief plus reference images to restyle, extend, or extract from.

Demo — Image to Image
Source image before editing with GPT Image 2.5
Input
Result after editing with GPT Image 2.5 Sunburst
Output — GPT Image 2.5 Sunburst

Flare vs Sunburst at a Glance

 FlareSunburst
Built forVolume and speed — social, thumbnails, product shots, prototypesFinished work — campaign creative, hero product imagery, print
QualityAbove GPT Image 2Highest in the family; cleaner detail and fewer artifacts
SpeedAbout half the latency of GPT Image 2Slower; trades time for quality
EditingSolid image-to-image and restylingTightest control across multi-step edits; preserves what you did not ask to change
Output1K / 2K / 4K, 13 aspect ratios, transparent background1K / 2K / 4K, 13 aspect ratios, transparent background
Credits per imageSame as Sunburst — priced by resolution, not by modelSame as Flare — priced by resolution, not by model

What You Get With GPT Image 2.5

Prompts are followed literally

GPT Image 2.5 fills the glass to the level you asked for, sets the clock to the time you wrote, and puts the details you listed in the frame. Fewer re-rolls to get the composition the brief actually described.

Text that reads

Headlines, labels, price tags, and packaging copy render legibly and spelled as written — the difference between a usable ad mock and a placeholder.

Edits that keep the rest intact

Upload up to four reference images and describe the change. The model swaps the background, restyles the subject, or extracts it onto transparency while leaving everything else where it was — Sunburst is the stricter of the two here.

Up to 4K, thirteen ratios

Square for the grid, 2:3 and 3:2 for print, 16:9 and 21:9 for banners, 9:16 for stories, plus 27:16, 16:27, 9:8, and 8:9 at 1K. Pick 1K to iterate and 4K to deliver.

Transparent backgrounds on demand

Set Background to Transparent to get a cut-out PNG straight from the model. At 2K and 4K describe an isolated subject with no backdrop or shadow so the alpha comes back clean.

How to Generate or Edit an Image in Three Steps

1

Choose Flare or Sunburst

Start with Flare while you are exploring — it is fast and costs the same per image. Switch to Sunburst for the version you will ship, or whenever an edit has to be exact.

2

Write the brief, or add images and describe the change

For text to image, name the subject, style, lighting, composition, and any text to render. For image to image, add one to four images and say what should change — the first image sets the composition, the rest act as references.

3

Set ratio, resolution, and background, then generate

Pick the aspect ratio your channel needs, 1K to 4K, and Auto, Opaque, or Transparent. The credit cost shows before you submit; the image appears with a download link when it is ready.

From Still to Motion

A GPT Image 2.5 render makes a clean first frame. Download it, then drop it into Image to Video to animate it with LTX 2.5, Seedance, MiniMax H3, or Kling — the composition, colours, and any rendered text carry into the clip. Generating the key visual and the motion in the same workspace keeps the look consistent across the whole set.

FAQ

What is the difference between GPT Image 2.5 Flare and Sunburst?+
Flare is the smaller, faster model — OpenAI positions it as the default, with image quality above GPT Image 2 at around 50% lower latency. Sunburst is the larger model optimized for quality and for precise control across edits, intended for production-ready campaign creative and polished product imagery. Both accept the same inputs and offer the same resolutions.
What resolutions and aspect ratios are supported?+
1K, 2K, and 4K. Aspect ratios: Auto, 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9, 27:16, 16:27, 9:8, and 8:9. The last four render at 1K only; everything else goes up to 4K.
Can GPT Image 2.5 edit my own image?+
Yes. Choose Image to Image, upload up to four images, and describe the change — restyle, replace the background, extend the canvas, extract the subject, or add text. The output resolution and ratio follow your settings, or match the source when set to Auto.
How do I get a transparent background?+
Set Background to Transparent. For 2K and 4K, the docs require the prompt to describe an isolated subject with no backdrop, scenery, or shadow; for image to image, explicitly ask to extract the subject and keep the background transparent.
Does Sunburst cost more than Flare?+
No. Both are billed per image by resolution — 1K, 2K, or 4K — so choosing Sunburst is a quality decision, not a budget one. The exact credit cost appears on the Generate button before you commit.
Are generated images moderated?+
Yes. Prompts and uploaded reference images pass through the same input moderation as the video tools, and prohibited sexual or NSFW content is blocked before generation.

Make the Image With GPT Image 2.5

Draft on Flare, finish on Sunburst. Pick the ratio, pick the resolution, and download a file that is ready to post, list, or print.

View Pricing and Credits