GPT Image 2.5

Sample collage showing GPT 2.5 editor layout for object, style, and background changes, not confirmed model output — Sample editing layout

Session history

Images generated in this session will appear here.

Continue your Image AI workflow

Quick answer

GPT Image 2.5 Editor

A GPT Image 2.5 editor changes an existing still with a written instruction and one or more reference images. This independent image-to-image workbench uses GPT Image 2.5 Sunburst by default for precise edits, with Flare available when you want a faster pass. Upload a JPG, PNG, or WebP, describe the change, then review the result before download. It is not ChatGPT and does not include Sketch or on-image comments.

The example stills show GPT Image 2.5 editor layout only. They are not confirmed GPT Image 2.5 outputs.

Published September 10, 2026 · Last updated September 10, 2026

Try the editor

Tool details

Primary workflow: GPT Image 2.5 editor from a reference image plus prompt
Default model: GPT Image 2.5 Sunburst, with Flare available in the same picker
References: Up to 16 images when the selected model allows it
Last updated: September 10, 2026

Examples

GPT Image 2.5 editor examples

Each card pairs a visible edit prompt with the intended change and a sample still. The images show workbench layout only; they are not claimed as live GPT Image 2.5 outputs.

Sample object-edit still used to show GPT Image 2.5 editor layout, not a confirmed GPT Image 2.5 output — Object change on a product still
Source
Product still used as a workbench layout sample
Instruction
Replace the speaker grille with a brushed-metal mesh, keep the bottle shape, lighting, and studio background unchanged.
Setting
Sunburst, one reference, 2K
A GPT Image 2.5 editor prompt should name what stays and what changes. This still is a workbench layout sample, not a confirmed GPT Image 2.5 output.

Sample portrait used to explain a GPT Image 2.5 editor background change — Background replacement on a portrait
Source
Portrait still used as a workbench layout sample
Instruction
Replace the background with a clean modern studio, keep the person, clothing, and face unchanged, preserve realistic edges.
Setting
Sunburst, one reference, 1K
Use the GPT Image 2.5 editor when the subject is already right and only the scene needs to move. Inspect hair edges and shadows after the run. This image is a sample layout.

Sample signage still used to illustrate GPT Image 2.5 editor text replacement — Visible text replacement
Source
Signage still used as a workbench layout sample
Instruction
Replace the visible headline with "Autumn Drop" while matching font weight, perspective, lighting, and material texture.
Setting
Sunburst, one reference, 2K
The GPT Image 2.5 editor can retarget short on-image words when the prompt quotes the replacement exactly. Always read the result; this sample is not a confirmed model output.

Core controls

What a GPT Image 2.5 editor can do

Keep the reference, instruction, Sunburst or Flare, ratio, and preview in one GPT Image 2.5 editor workspace before you spend credits.

Edit from a reference instead of starting over

Keep identity, product shape, or layout while changing one local part of the still.

The GPT Image 2.5 editor accepts a prompt plus up to 16 reference images and returns a revised still in the same workbench.

Input: a source image and a targeted instruction. Output: an edited still at 1K, 2K, or 4K.

Sunburst-first precision, with Flare for faster passes

Default to the precision tier when edges, labels, or likeness matter. Switch to Flare only for a faster look at the direction.

Sunburst is the default GPT Image 2.5 editor model. Flare uses the same 25, 40, or 65 credit table and is better for a quick exploratory edit.

Input: Sunburst or Flare plus ratio and resolution. Output: one edited still.

Natural-language object, background, and text changes

Describe the edit in words instead of masking by hand. Say what must stay as clearly as what must change.

This GPT Image 2.5 editor is built for image-to-image jobs: object swap, background replace, short text retarget, and multi-reference merges.

Input: a keep/change instruction plus the source file. Output: a revised composition to inspect.

Review original and edited frames together

Compare the source with the new still before you download or send the result to upscale or video.

After a GPT Image 2.5 editor run, use History and the original/edited comparison when a single reference is present.

Input: a completed edit. Output: a downloadable still you can later upscale or animate.

How to

How to edit an image with GPT Image 2.5

Upload a source you have rights to, write one targeted change, then compare the original and the edit.

1
Upload the source still
Add a JPG, PNG, or WebP you have the right to edit. The GPT Image 2.5 editor works best when the main subject is already clear in the source.
2
Write a targeted change
Name what must stay and what must change. Quote replacement text exactly. Avoid stacking many unrelated edits in one prompt.
3
Choose Sunburst or Flare and generate
Keep Sunburst for precise image-to-image work. Use Flare for a faster look. Pick ratio and 1K/2K/4K with the same Auto and 1:1 constraints as generation.
4
Inspect, download, or continue
Check edges, identity, and lettering. Download the edit, upscale it, or animate it. Sign-in is required before credits are spent.

Settings

Model, references, and resolution for this edit

Use this table to pick a GPT Image 2.5 editor model, reference count, and resolution that match the job.

Setting	Available options	Best for
Model	GPT Image 2.5 Sunburst or GPT Image 2.5 Flare	Sunburst for a GPT Image 2.5 editor pass that must protect product edges or labels. Flare for a faster exploratory edit.
Reference images	1–16 stills, depending on the selected model	One source for a local change. Extra references when you need to merge a subject into another scene.
Prompt	Text instruction, up to 20,000 characters	Keep/change language, quoted replacement text, and lighting or material constraints.
Resolution and credits	1K / 25 credits, 2K / 40 credits, 4K / 65 credits	Test the instruction at 1K, then raise resolution once identity and layout look right.

Use cases

Typical image-to-image jobs

Use this workflow when a still already exists and you need a local change before upscale, animation, or download.

Ecommerce teams swapping product details

Starting input: A pack shot plus a short instruction to change a label, colorway, or prop while keeping the bottle or box.
Tool action: Run the GPT Image 2.5 editor on Sunburst, quote the new label text, and inspect edges before download.
Expected result: A revised product still ready for a listing mock or a second upscale pass.

Photographers replacing a background

Starting input: A portrait or lifestyle frame where the subject is good and the scene is not.
Tool action: Upload the still, describe the new environment, and tell the GPT Image 2.5 editor to keep face, clothing, and lighting.
Expected result: A background-changed draft that still needs a human check on hair and contact shadows.

Designers retargeting campaign lettering

Starting input: A poster or thumbnail whose layout is right but whose headline must change.
Tool action: Quote the replacement words and ask the model to match perspective and material. Prefer Sunburst.
Expected result: A text-updated still. Read every character before ads or client review.

Limits

Limits of this independent editor

Know the credit cost of this GPT Image 2.5 editor, missing ChatGPT canvas markup tools, and review checks before you treat an edit as production art.

This is not ChatGPT

The GPT Image 2.5 editor is an independent image-to-image workbench. It is not the ChatGPT app, not ChatGPT Images 2.5, and not affiliated with OpenAI.

No Sketch canvas or on-image comments

This tool does not expose ChatGPT canvas markup, click-to-comment regions, or official templates. Put the edit in the prompt, or attach extra references instead.

Credits and rights

Generation spends 25, 40, or 65 site credits for 1K, 2K, or 4K. Only upload images you have the right to edit. Avoid protected likenesses, trademarks, and private photos.

Quality control

Complex crowds, tiny UI, detailed hands, and long text blocks can drift. Compare the original and the edit before publishing or paid media.

FAQ

GPT Image 2.5 editor FAQ

Direct answers about Sunburst, references, on-image text, credits, and how this tool differs from ChatGPT.

What is a GPT Image 2.5 editor?

A GPT Image 2.5 editor applies a written instruction to an uploaded still using GPT Image 2.5 Sunburst or Flare. It is an independent image-to-image workbench, not the ChatGPT editor.

Should I use Sunburst or Flare for editing?

Use Sunburst for most GPT Image 2.5 editor jobs that must keep identity, labels, or product edges. Use Flare when you only need a fast look at the direction. Both tiers cost the same credits here.

How many reference images can I upload?

The GPT Image 2.5 editor supports up to 16 reference images. Start with one clear source for a local change, then add more only when you need to merge a subject into another scene.

Can I replace text inside an image?

A GPT Image 2.5 editor can replace short on-image text when the prompt quotes the new words and asks the model to match font, perspective, and lighting. Always read the result; small lettering can still drift.

Is this a GPT Image 2.5 photo editor for free?

You can open the workbench and stage an upload without paying. Running the edit requires sign-in and site credits. The GPT Image 2.5 editor does not offer an unlimited no-credit mode.

Next steps

After you finish an edit

Continue from this GPT Image 2.5 editor into a new still, an upscale, or image-to-video.

GPT Image 2.5 generatorStart a new still from a written prompt when you have no source.AI image upscalerEnlarge a finished edit at 2x or 4x and check 4K readiness.Image to videoAnimate the edited still as a start frame.

Ready to edit

Continue in the GPT Image 2.5 editor

Upload a source still, write one targeted change, then review the result before you upscale, animate, or publish.

Edit an image Generate a new still Upscale a finished still