Skip to main content
LogoClipDance
NewFlare and Sunburst are both liveOpen the editor

GPT Image 2.5, no API key and no setup

GPT Image 2.5 is OpenAI's newest image model, and this page runs it in a browser tab. Type a prompt or drop in up to 16 references, pick a quality tier, and download a finished frame in seconds.

What is GPT Image 2.5?

GPT Image 2.5 is the image generation and editing model OpenAI shipped on 8 September 2026, in two versions that read exactly the same settings. Flare renders at roughly half the latency of the previous generation and is the one to reach for most days. Sunburst takes longer and holds detail together across a long chain of edits. Both draw text you can actually read, keep faces from your reference photos recognizable, and change only the part of a picture you asked about. On this page GPT Image 2.5 runs on our credits, so there is no OpenAI account, no billing setup and no key to rotate.

Reference images per edit, plus an optional mask
16
Quality tiers, two of them new above High
5
Up to 8.3 megapixels, any shape between 1:3 and 3:1
4K

Every GPT Image 2.5 render is billed by the frame it produces. A 1K draft is 1 credit; the ceiling is reserved for the tiers and sizes that actually need it.

Studio portrait rendered by GPT Image 2.5 Sunburst at 1776 by 2368 pixels, with individual hair strands, skin texture and the weave of the wool coat held intact

GPT Image 2.5 edits stop where you told them to

The reason to reach for GPT Image 2.5 over a general image model is restraint. Name one object and the rest of the frame survives untouched, including the parts that are hard to regenerate: skin texture, brand type, the grain in a background wall. Chain five edits together and the first four still hold.

Start editingSee credit pricing

Text to image

Describe the frame and pick a shape. Anything from a square avatar to a 3840-pixel banner.

Reference editing

Drop in up to 16 photos. Faces, products and materials come back recognizable rather than reinterpreted.

Masked inpainting

Paint over the region you want replaced and leave everything outside it exactly as shot.

Batches of ten

Ask GPT Image 2.5 for up to 10 variations in one request and keep the one that works.

Made with GPT Image 2.5 on this page

Every frame below came out of the generator above, at the model, tier and size printed on the card. No retouching and no upscaling pass.

Editorial studio portrait of a woman in a charcoal wool coat and gold silk scarf, lit by a single soft key light against black
Sunburstxhigh1776×23685 cr
Profile portrait of a bearded man in a black roll-neck, one hard warm key light carving the near side of his face
Sunburstxhigh1776×23685 cr
Matte-black ceramic perfume bottle with a brushed-gold collar on dark marble, lit by a single soft window light
Sunbursthigh2048×20484 cr
Ultra-wide 21:9 panorama of black volcanic sand dunes under a low amber sun
Flarehigh3136×13442 cr
Vertical editorial fashion frame of a gold silk coat rim-lit against a black corridor
Sunbursthigh1536×27362 cr
Architectural interior of a walnut and blackened-steel library at dusk with light beams through dust
Flarehigh2736×15362 cr
Overhead frame of saffron risotto in a dark-glazed bowl on a charred wooden table
Flarehigh2048×20484 cr
Vintage brass desk lamp with a green glass shade, rendered on a transparent background as a clean cutout
Flarehigh2048×20484 cr

Why run GPT Image 2.5 here

The model is the same one OpenAI serves. What changes is everything around it.

Using GPT Image 2.5 directly means an OpenAI account, a funded balance, and a key you have to keep out of your repo. Here you sign in and generate. New accounts start with free credits, so the first few frames cost nothing.

Both GPT Image 2.5 versions sit in the model list and read identical settings, so comparing them is a dropdown rather than a code change. Draft on Flare while you are still deciding, then re-run the winner on Sunburst when the edit has to survive scrutiny.

The credit figure updates as you change tier, size and batch count, so an expensive combination announces itself before you commit. Failed generations are refunded automatically, and your renders stay in your library instead of expiring behind a signed URL.
Matte-black ceramic bottle with a brushed-gold collar, rendered by GPT Image 2.5 Sunburst at 2048 by 2048 pixels

What people build with GPT Image 2.5

Six jobs it handles without a round trip through Photoshop.

Product shots

Put the same bottle on six backgrounds. The label stays readable and the lighting follows the new scene.

Cutouts with real transparency

Ask for a transparent background and get a PNG or WebP with a usable alpha channel, ready to drop onto any layout.

Posters with legible type

Headline text lands as text, not as the melted lettering earlier image models produced. Set a 3:1 frame and it still holds.

A character across a series

Feed the same face in as a reference on every render and the series reads as one person rather than five cousins.

Targeted retouching

Remove the cable, change the shirt colour, swap the sky. A mask confines the edit and the rest of the pixels are left alone.

One asset, every aspect ratio

Re-render the same idea at 9:16, 1:1 and 16:9 instead of cropping and hoping the composition survives.

How to use GPT Image 2.5

Four steps, and the third one is where the money is.

1

Write the prompt

Say what should be in the frame. For an edit, name only what changes and leave the rest unsaid.

2

Add references

Optional. Up to 16 images, plus a mask if the edit needs to be confined to one region.

3

Pick tier and size

Low for thumbnails and layout tests, High for anything client-facing, Max when the detail is the whole point.

4

Generate and download

Frames arrive in seconds and stay in your library as PNG, JPEG or WebP.

GPT Image 2.5 questions

Short answers, and the numbers behind them.

They take exactly the same settings and cost the same here. Flare is the small, fast version and is the sensible default for drafts, social assets and anything high volume. Sunburst is the larger one and holds detail together more reliably across a long chain of successive edits, which is what you want for campaign work and finished product imagery. Switching between them is one dropdown, so the practical approach is to draft on Flare and finish on Sunburst.

There is no single per-image price, because GPT Image 2.5 is metered by the frame it renders. A 1K image at Low or Medium is 1 credit. High at 1K sits between 1 and 2 credits depending on shape, and the Max tier at 4K reaches the low twenties. The generator shows the exact figure before you press the button, and a failed render is refunded automatically.

No. That is the point of this page. Going to OpenAI directly means an account, a funded balance and a key to manage. Here you sign in with an email and spend credits. New accounts get free credits on signup, which covers a first batch of real prompts.

Yes, and it is the strongest thing the model does. Upload between one and sixteen reference images, describe the change, and the parts you did not mention stay as they were. Add a mask and the edit is confined to the region you painted. Faces and product details from the references come back recognizable rather than redrawn from scratch.

Anything from roughly 0.65 to 8.3 megapixels, with both edges a multiple of 16, neither edge over 3840 pixels, and an aspect ratio between 1:3 and 3:1. In practice that covers every social and print shape people ask for, including 21:9 banners and 9:16 vertical. Pick a ratio and a 1K, 2K or 4K tier and the exact pixel size is worked out for you.

Yes. Set Background to Transparent and choose PNG or WebP as the output format, and the file arrives with a real alpha channel. JPEG cannot carry transparency, so it is disabled while that setting is on. This is the fastest route to a clean cutout without a separate background-removal pass.

Up to 10 per request. Each rendered frame is billed, but the prompt and any reference images are charged once for the whole batch rather than once per image, so a batch of ten costs noticeably less than ten separate requests at the same settings.

Yes. Images you generate are yours to publish, sell or build into client work, subject to OpenAI's usage policies on the content itself. They stay in your library rather than behind an expiring link, so you can come back for the source file months later.

Generate your first GPT Image 2.5 frame

Free credits on signup. No API key, no card, no setup. Pick a tier and start.

Open the generatorSee pricing