By the end of this walkthrough you will have one still from the GPT Image 2.5 generator, plus one local edit that did not redraw the subject — then a file you can download into Smart Canvas or Multi-Size.
How to use GPT Image 2.5 on KreadoAI is not a ChatGPT consumer-canvas lesson. It is a three-step SGE path: Select the model, Generate from a prompt or a reference, Edit one named change, then ship. According to OpenAI’s September 8, 2026 launch, ChatGPT Images 2.5 (GPT Image 2.5) is built for sharper detail, tighter local edits, and stronger subject lock from reference photos. Those gains only show up if you actually pick the model, set ratio before you hit generate, and stop rewriting the whole scene on the edit turn.
This gpt image 2.5 tutorial follows the same three steps as GPT Image 2.5 AI Image Generator. For the brief language itself, use the GPT Image 2.5 prompt guide. This article is the click path.
What GPT Image 2.5 Does Well on KreadoAI
We ran SGE the same week as the OpenAI launch: a scale still, a quoted-type social still, and a one-change identity edit. The model earns the credit when you treat it as generate-then-edit, not as a one-shot art button.
Product stills that keep the pack. Upload a white-background or packaging photo, lock shape and color, then change only the surface or the light. That is the ecommerce job — hero, lifestyle, detail — without inventing a second bottle.
Precise image editing. Change only the background, prop, or color you name. Keep the person, clothing, pose, or bottle geometry so review can use the file. One change per turn. If a later turn drifts, paste the lock list again.
Marketing stills with readable type. A 1:1 or 9:16 creative can carry a headline and a layout, not just a pretty product. Traffic it toward Meta or Instagram Ad Maker sizes after you QA spelling.
Identity-locked scenes. Keep face and outfit, move the same person into a new street or studio. Same logic as the pack lock — assign the reference a role before you ask for a new set.
Scale is an Elements instruction. If the figure is the same size as the can, the brief never named the lock.
What GPT Image 2.5 Can't Do in This Workflow
SGE is a KreadoAI path. It is not a dump of every ChatGPT consumer feature, and it is not a developer console.
| You wanted | What happens on KreadoAI | What to do |
|---|---|---|
| Draw a layout by hand inside ChatGPT | That canvas is not this generator | Describe layout in the prompt, or composite in Smart Canvas |
| Paste API model names into the box | Ratio and resolution are UI controls | Set size in the generator, repeat the ratio in the brief |
| Three edits in one turn | Face, label, or crop often drift | One named change. Then another turn |
| Pixel-identical legal label | Prompt lock is close, not a flatten | Stop prompting; composite the real label in Smart Canvas |
| Offer-first 2.0 statics you already traffic | 2.5 is a different model slot | Keep GPT Image 2 selected if that campaign is live |
OpenAI’s own framing is “more reliable,” not “skip QA.” After generate, check spelling, extra logos, and whether the pack still matches the reference.
How to Use GPT Image 2.5 on KreadoAI: 3 Steps
SGE maps 1:1 to the product page. Do not add a fourth “design the ad in a new novel” step between generate and edit — that is how hierarchy collapses.
Step 1 — Select GPT Image 2.5
Open image generation. In the model list, choose GPT Image 2.5. Confirm you are not still on GPT Image 2, Nano Banana, or Seedream from the last session.
Expected result: a prompt box, a reference-upload control, and size / resolution fields. If those three are visible, Select is done. If you skip this click, the rest of the tutorial is a different model.
Deep-link when you can: GPT Image 2.5 on KreadoAI already has the model query on the URL.
Step 2 — Prompt or reference, then generate
Paste a brief, or upload a reference, or both. Set ratio before you generate so you are not fighting a square crop later.
| Output | Ratio to set | Notes |
|---|---|---|
| Feed / square PDP | 1:1 | Keep headlines off the product center |
| Stories / Reels companion | 9:16 | Leave top and bottom safe for UI chrome |
| Blog / landscape hero | 16:9 | Leave a quiet third if type comes later |
If you uploaded a pack or a face, say what that file owns: identity, product geometry, or style only. Then generate.
Expected result: a first still that already matches the ratio and the subject placement. If the tiny figure comes back life-size, you skipped the scale lock — do not “fix it” by rewriting the whole picnic. Go to Step 3 with a Change / Preserve pair, or generate once more with a shorter lock list.
Generate is a full still. If the file is an ad, quote the headline in this box — not in a later novel.
How to generate images with gpt image 2.5 without a reference: write the asset type, the subject, the light, and the exclusions. How to generate with a reference: upload first, then name the lock. Longer KEEP templates live in the prompt guide; this step only needs the file to exist.
Step 3 — Local edit, download, Smart Canvas
Open local edit on the still you just generated (or upload an existing frame). Write one change. List what must not move. Match lighting on the new piece.
Example we ran: change only the background from an overcast plaza to a rainy neon night street. Preserve face, hair, overshirt, pose, crop, camera. Relight the jacket. Do not redesign the person.
Expected result: the same man in a new city. If you got a new face, the Preserve list was implied, not written.
Download the file. Then pick the next surface:
- Smart Canvas when you need layers, type overlay, or a real label pasted back
- Multi-Size Ad Image Generator when the winner needs 1:1, 4:5, and 9:16 from one hero
Do not stay in the generator hoping a fifth rewrite will typeset a legal disclaimer. That is a canvas job.
GPT Image 2.5 Examples We Actually Ran
Three stills, same week as the launch, same SGE order. Swap the nouns; keep the run order.
Generate — scale still. Kind: 16:9 cinematic still. Elements: palm-sized figure in a yellow raincoat on the rim of a giant unlabeled citrus can; lime like a hill; ice like boulders; picnic table as landscape; golden hour. Exact copy: none. Preserve: no extra people, no brand letters, no watermarks. Use this when the brief is a complex visual instruction, not a stock pack shot.
Local edit — identity lock. Change only the background to a rainy neon night street. Preserve face, hair, overshirt, pose, crop, camera. Relight to match neon. Use this when you already have the person and need a new set without a reshoot.
Quoted type — social still. Kind: 16:9 paid-social still. Elements: matte black can, wet stone, tropical water, citrus, splash. Exact copy: headline "SUMMER ENERGY" once; banner "Refresh Your Day" once; no other words. Use this when the asset is the ad, not a blank beauty frame. Full KEEP strings are in the prompt guide; here the point is that type belongs in the first generate.
If the face changed, you rewrote the person. If only the city changed, SGE held.
Who it is for: ecommerce teams that need reference-true pack shots; performance teams that need a 1:1 with readable type; anyone who already has a face or SKU photo and needs a new set without a new shoot.
FAQ
How do I use GPT Image 2.5 on KreadoAI?
Select GPT Image 2.5 in image generation, paste a prompt or upload a reference, set ratio, generate, then run one local edit if needed. Download, then continue in Smart Canvas or Multi-Size. That is how to use gpt image 2.5 on this site.
Where do I select GPT Image 2.5?
Open the image generator and pick GPT Image 2.5 in the model list, or use the model link on GPT Image 2.5 on KreadoAI. Confirm you are not still on GPT Image 2.
Can I upload a reference image?
Yes. Upload before you generate. Tell the brief whether the file owns identity, product geometry, or style. A reference with no role is just another mood board.
How does GPT Image 2.5 image editing work?
Write “Change only [X]. Preserve [Y].” One change per turn. Relight the new element so it sits in the old crop. If legal needs an identical label, stop and composite in Smart Canvas.
Do I need the prompt guide before I start?
No. SGE will get you a first file. Use the GPT Image 2.5 prompt guide when the still is pretty but unusable — missing Kind, Exact copy, or Preserve.
GPT Image 2.5 vs GPT Image 2 — which should I open?
Open 2.5 for tighter local edits and lock-from-reference stills. Stay on 2.0 if that campaign already ships offer-first statics. Both remain selectable; do not migrate a live 2.0 brief blindly.
What size should I generate?
Set the ratio you will traffic: 1:1 feed, 9:16 stories, 16:9 landscape. Repeat it in the brief. Cropping a square into a story after the fact is how products lose their heads.
Is GPT Image 2.5 good for ecommerce and marketing?
Yes, when the job is a pack that must stay the pack, a person that must stay the person, or a poster with quoted type. It is a generator and an editor on KreadoAI, not a substitute for platform trafficking QA.
