GPT Image 2.5 features, one by one
Nine changes ship with this release. Some are model behaviour, some are ChatGPT interface work, and the difference matters when you decide where to run a job.
A version number ending in .5 usually signals refinement rather than reinvention, and that is exactly what this is. Nothing in GPT Image 2.5 does a category of work the previous model could not attempt. What changed is the reliability of the attempt — and for production use, reliability is the entire product. An editor that gets the jacket right nine times in ten is a tool. One that gets it right five times in ten is a slot machine.
The nine changes below split cleanly in two. Six are model behaviour and travel with the API models, so a developer building on GPT Image 2.5 gets them. Three are ChatGPT interface features that do not exist outside the chat app. Teams routinely plan a pipeline around Sketch or templates and then discover there is no endpoint for them, so the split is worth internalising before you design anything on GPT Image 2.5.
Model behaviour
Available through both ChatGPT and the API.
Reference-photo fidelity
Working from a reference photo, subjects stay recognisable across new settings, styles and compositions. Lighting and texture read more naturally, and distinctive features are more likely to survive the transfer. This is the change OpenAI leads with, and it is the one that matters for anyone putting a real person, pet or product into a generated scene.
Scoped editing
Ask for one change and you get one change. The model is better at leaving untouched regions alone, even when subject and background are visually complex — swap a jacket while the pose holds, replace a backdrop while the product geometry stays put, rewrite copy without the layout shifting underneath it.
Edits that survive the thread
In long conversations, earlier edits carry forward instead of quietly decaying. Each new instruction builds on the last rather than re-rolling the image, which is what makes a ten-turn retouching session viable instead of a race against drift.
Up to 50% lower latency
Generation latency drops by up to half against Images 2.0. Autoregressive image generation has been the slow part of this model family since the beginning, so halving it changes the interaction pattern: you iterate rather than wait, submit, and go do something else.
Complex layouts and transparency
Denser layouts hold together, and transparent-background generation improved enough that Manus called it out specifically. For anyone producing overlay assets, sticker sheets or UI art, a clean alpha channel out of the box removes an entire cutout step.
Grounded real-world content
Images that contain real-world information render more accurately, which shows up most in infographics, charts, maps and anything with a factual label on it. The system card also flags improved infographic accuracy and layout as a headline capability.
ChatGPT-only features
Interface work. There is no API equivalent for any of these.
Sketch
Type @Sketch in ChatGPT and draw directly in the app. The drawing becomes a reference for generation, which is the fastest route for anything easier to draw than describe — a layout, a pose, a rough spatial arrangement of objects.
Templates
Starting scaffolds for common formats such as posters and merch. You pick a template, then add your own message, design elements and style. It is a cold-start fix rather than a model capability, and it removes the blank-prompt problem for people who do not write prompts for a living.
Shareable prompts
A generated image can carry the exact prompt that produced it, so someone else can run the same idea against their own photos. It turns a finished image into a reusable recipe instead of a dead end.
Why scoped editing is the real story
Every earlier model in this family regenerated the picture on every turn. Ask for a different jacket colour and you did not get the same photograph with a different jacket — you got a new photograph that resembled the old one, with a different jacket and also a slightly different face, a shifted crop, and a background that had quietly rearranged itself. Practitioners called it drift, and it made long edit chains unusable. GPT Image 2.5 is the first release in this family to treat that as a defect rather than a property of the medium.
The fix OpenAI describes is preservation: change the elements named in the instruction, leave the rest intact, and do so even when subject and background are visually complex. Higgsfield AI, one of the launch partners, described the same behaviour as preserving an original image's character, composition and visual identity through edits. That is the language of a retouching tool, not a generator, and it is the clearest signal of where this release is aimed.
The companion change is persistence. In a long conversation, earlier edits carry through, and each new instruction builds on previous work rather than degrading it. Together the two turn a chat thread into an edit history. That has a practical consequence worth stating plainly: when an edit goes wrong, ask for the previous state back instead of starting a new thread. Restarting throws away every refinement you already paid for.
The latency change, in workflow terms
Up to fifty percent lower generation latency than Images 2.0 is the number OpenAI publishes. Manus, evaluating the Flare tier independently, reported two to four times the speed of GPT Image 2 on its own test set. The gap between those two claims is ordinary — vendor figures average across a distribution, partner figures reflect one workload — and both point the same way.
What that buys is a different interaction pattern. Under the old timing you front-loaded everything into one enormous prompt because a second attempt was expensive in wall-clock terms. Now the cheap move is a short prompt followed by three corrections, which suits the scoped-edit behaviour perfectly. The speed change and the editing change reinforce each other in GPT Image 2.5; neither is half as useful alone.
What launch partners reported
“The GPT-Image-2.5 models are available in Adobe Firefly, Adobe's creative AI studio.”
“Flare delivered high-quality images at two to four times the speed of GPT Image 2, with improved transparent-background generation.”
“The model preserves an original image's character, composition and visual identity through edits.”