1
Write the prompt as a shot list
Name the subject, the light, the framing and the finish. The parts you leave out get invented — that is as true of 2.5 as of every model before it.
Free · no sign-up · runs in your browser
GPT Image 2.5, and a free generator you can run right now
OpenAI announced GPT Image 2.5 on 8 September 2026 — sharper detail, more precise editing, Sketch in ChatGPT, and up to 50% lower generation latency. This page tracks exactly what shipped. While it reaches the providers, run the same jobs — text rendering, reference photos, instructed edits — free on GPT Image 2 in the generator below: in your browser, no sign-up.
Two API models50% lower latencyGenerate free below
2.5 output
What GPT Image 2.5 actually produces
Four pictures posted publicly on X the day it shipped, by the people who made them — not runs from this workspace, and not our generator's output. They are here because they show what the model does; the generator below runs GPT Image 2.
Lettering
Carved Latin, held on stone
What GPT Image 2.5 is
GPT Image 2.5 is the image generation and editing model OpenAI announced on 8 September 2026. It carries two names: ChatGPT Images 2.5 in the product, GPT Image 2.5 in the API. It replaces Images 2.0.
What it does
GPT Image 2.5 is not on the providers this site buys from yet, so the generator on this page runs GPT Image 2 — the model that shipped in June and still handles text rendering, reference photos and instructed edits. When 2.5 reaches a provider, it goes into the picker.
Every capability claim on this page traces to OpenAI's launch announcement and its developer post. Numbers that OpenAI did not publish — resolution ceilings, exact per-image prices by size — are left out rather than guessed at.
Creative engine
GPT Image 2.5 at a glance
The facts OpenAI published on launch day, and what this site runs while providers catch up.
8 Sep 2026
Announced
Flare · Sunburst
API models
GPT Image 2
Running here today
Two names, one model line
The API exposes it as two models. Flare is the default high-volume path and carries the quality, editing and speed gains. Sunburst adds an extra level of precision for detailed creative work, spending longer on generation to get it.
Editing got the attention
OpenAI's launch notes lead on localised editing that leaves surrounding detail intact, better subject preservation from reference photos, and steadier consistency across repeated edits. That is where 2.5 differs most from what came before.
Sketch is a ChatGPT feature
Sketch lets you draw inside ChatGPT and hand the drawing to the model as a visual guide. It is part of the ChatGPT product surface, not a separate API parameter.
The changes
What changed in GPT Image 2.5
Four claims came with the launch. Each one below is OpenAI's own, stated as they stated it.
Sharper detail, richer texture
More natural lighting and richer surface texture than Images 2.0. The difference shows first on skin, fabric and metal, where flat rendering used to give the output away.
Up to 50% lower latency
Generation latency is up to 50% lower than Images 2.0. On an image model that is felt rather than measured: it changes whether you iterate on a prompt or give up on it.
Instructions followed more reliably
Edits land where you asked and leave the rest alone. Subject preservation from a reference photo improved, which is the failure everyone has hit — the face comes back slightly different.
Sketch as an input
Draw a rough composition in ChatGPT and the model treats it as a layout guide. Composition control by drawing rather than by describing.
Flare and Sunburst
GPT Image 2.5: Flare and Sunburst
The API split is a speed-against-precision choice, and OpenAI named both rather than hiding them behind a quality flag.
Flare — the default
The high-volume path. It carries the quality, editing and speed improvements and is the one to reach for when a batch has to come back quickly.
Sunburst — the precise one
An extra level of precision for detailed creative work. Reach for it on the frame that has to be right, and Flare for the fifty that have to be quick.
Both are API-side
Flare and Sunburst are how 2.5 appears to developers. Inside ChatGPT the split is not exposed — you get the model, not the choice.
What OpenAI did not publish
Per-image prices by size, resolution ceilings and rate limits were not part of the announcement. This section gets updated when OpenAI publishes them.
How to use it
How to generate while GPT Image 2.5 reaches the providers
Three steps, no account. The generator on this page is open to anyone who lands on it.
2
Generate free in the browser
Press generate. The free engine runs on your own daily allowance, in your browser, without an account.
3
Edit rather than re-roll
When one result is close, change it by instruction instead of regenerating from scratch. That is the workflow 2.5 sharpened, and it already works on the model running here.
What people bring here
The jobs people came to GPT Image 2.5 for
What the launch coverage pointed at, and where each one lands on this site today.
01
Posters and layouts with real type
Legible words inside a picture is the capability that separates this model line. Run it on the generator above, or start from a written prompt in the prompt library.
02
Editing one thing in a photo
Localised edits that leave the rest of the frame untouched — the headline improvement in 2.5, and the job the ChatGPT image editor already does here.
03
Keeping a character consistent
The same face across several frames, held by naming the identity anchors in every prompt and editing from the first result rather than starting over.
04
Product and campaign frames
Studio light, a clean surface and a legible logo. When the background has to go, the background remover takes it off without clipping the edges.
05
Making a keeper bigger
When a frame comes out exactly right, you want it bigger. The image upscaler enlarges it without re-generating and losing what made it good.
Answers
GPT Image 2.5 questions
Yes. OpenAI announced it on 8 September 2026 — branded ChatGPT Images 2.5 in the product and GPT Image 2.5 in the API. Pages still describing it as a rumour were written before that date.
Not yet. The generator here runs GPT Image 2 today — free, in your browser — and GPT Image 2.5 appears in the picker as soon as a provider carries it.
Flare is the default: the same quality and editing gains, built for volume and speed. Sunburst spends longer on generation to reach an extra level of precision for detailed work. Both are API models; ChatGPT does not expose the choice.
OpenAI has not published per-image prices by size yet, so this page will not invent one — it gets updated when the official numbers land. What you can price today is a run here: the credit cost appears before you generate, and a failed run is refunded.
In ChatGPT it follows the usual plan limits. On this page the generator is free and needs no account — it runs GPT Image 2 on a per-visitor daily allowance in your browser.
Sketch is a ChatGPT feature that lets you draw a rough composition and pass it to the model as a guide. It is a way to control layout by drawing instead of describing. A prompt that names framing does much of the same work, and Sketch gives you that control by drawing instead of describing.
No. Prompting does not change between versions: the wording that works on GPT Image 2 is the wording that will work on 2.5. Anything you write now carries over.
One credit balance for every image model
Credits are shared across everything Veida AI runs. You see the cost before each run, and a failed job is never charged.
Free · 1 free image now, no account
Plus unlimited runs on the free engine — no card, nothing to cancel.
Start hereIncl. tax
Plus
Enough for a steady week of drafting.
$59.99
$29.99 / month
2,500 credits
- 625 × Veida AI Image v1
- $0.048 each
- 50 × Nano Banana 2 4K
- $0.60 each
GPT Image 2Nano Banana 2Seedream 5.0 ProQwen Image 3
What's included
- 2 generations at once
- Every model on this page — GPT Image 2, Nano Banana 2, Seedream 5.0 Pro, Qwen Image 3
- Up to 4K output
Secure checkout by Stripe. Card details never touch our servers.
PopularIncl. tax
Pro
The pick if you generate most days.
$179.99
$89.99 / month
Unlimited creditsUnlimited images
- Full speed each month
- ~240 flagship 4K
- or on the fastest model
- ~3,000 images
GPT Image 2Nano Banana 2Seedream 5.0 ProQwen Image 3
What's included
- 4 generations at once
- Unlimited images — no credit cost, on any model
- Every model on this page — GPT Image 2, Nano Banana 2, Seedream 5.0 Pro, Qwen Image 3
Secure checkout by Stripe. Card details never touch our servers.
Best valueIncl. tax
Max
For people who generate every day.
$299.99
$149.99 / month
Unlimited creditsUnlimited images
- Full speed each month
- ~400 flagship 4K
- or on the fastest model
- ~5,000 images
GPT Image 2Nano Banana 2Seedream 5.0 ProQwen Image 3
What's included
- 8 generations at once
- Unlimited images — no credit cost, on any model
- Every model on this page — GPT Image 2, Nano Banana 2, Seedream 5.0 Pro, Qwen Image 3
Secure checkout by Stripe. Card details never touch our servers.
Not sure which to choose? Contact us
Charges appear as “VEIDA AI” on your card statement. You can cancel any time from Settings → Billing; cancellation takes effect at the end of the period you already paid for. Operator details, the full model list, and the refund window are on the about page.
Keep going
Where to go from GPT Image 2.5
The model running here today, and the tools built on top of it.
Generate before GPT Image 2.5 gets here
Get your first result free, in your browser, without creating an account. Write a prompt and see what comes back.
