OpenAI has updated ChatGPT’s image generator again, and this release focuses on some of the problems that matter once you move beyond making a single interesting picture.
ChatGPT Images 2.5 generates images faster, handles reference photos more faithfully, makes more controlled edits and does a better job of keeping an image consistent through repeated revisions. OpenAI has also added new ways to work visually, including Sketch, image comments, templates and shareable prompts.
The most noticeable change may not be raw image quality. It is control.
Anyone who has spent time editing AI-generated images knows the problem: you ask to change one small thing, and the tool quietly changes three other things you wanted to keep. Images 2.5 is designed to reduce that behavior.
OpenAI says generation latency has also been reduced by up to 50% compared with ChatGPT Images 2.0.
What is ChatGPT Images 2.5?
ChatGPT Images 2.5 is OpenAI’s latest image-generation and editing system inside ChatGPT. It can create images from text, transform uploaded photos, make targeted edits, work from sketches and maintain visual details across several rounds of editing.
It launched on September 8, 2026 and is rolling out across ChatGPT, ChatGPT Work and Codex on desktop, mobile and web. OpenAI says the image experience is available across ChatGPT tiers.
There is an important naming distinction.
ChatGPT Images 2.5 refers to the image experience people use inside ChatGPT. Developers using OpenAI’s API now have two related models called GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst.
ChatGPT Images 2.5 vs Images 2.0
The upgrade makes more sense when you look at what has changed rather than treating “2.5” as a simple quality bump.
| Capability | ChatGPT Images 2.0 | ChatGPT Images 2.5 |
|---|---|---|
| Image generation | Existing baseline | Up to 50% lower generation latency |
| Reference photos | Supported | Better subject and likeness preservation |
| Targeted editing | Supported | More precise changes with fewer unintended alterations |
| Multiple edits | Supported | Better consistency across successive revisions |
| Lighting and texture | Existing image quality | More natural lighting and richer textures |
| Complex layouts | Supported | Improved layout and visual-instruction handling |
| Transparent backgrounds | Supported | Improved handling |
| Sketch-to-image | Not part of the 2.0 launch | Added with Sketch |
| Image comments | Not part of the 2.0 launch | Added for focused editing |
| Templates | Not part of the 2.0 launch | Added for formats such as posters and merchandise |
| Prompt sharing | Basic image sharing | Prompts can also be shared for reuse |
| API models | Earlier GPT-Image models | Flare and Sunburst |
OpenAI introduced Images 2.0 in April 2026. Images 2.5 arrives less than five months later, so this feels less like a complete reset and more like an attempt to make AI image generation useful for longer creative workflows.
1. Editing is more precise
This is probably the most useful improvement.
With generative image editors, changing one object can inadvertently modify a person’s face, the lighting, background details or even text elsewhere in the image.
OpenAI says Images 2.5 is better at changing only the requested element while preserving the rest of the composition. That includes fairly complicated scenes and backgrounds.
This matters particularly for product images, advertisements, social creative and branded assets.
Imagine that you have finally generated a usable product photograph but want to change the color of the packaging. Regenerating the entire scene can introduce unwanted differences. A more reliable targeted-editing system makes iterative production considerably more practical.
Early independent testing reported by Axios also found improved likeness preservation when transforming existing photographs, although, as with any generative editor, users should not assume that untouched portions of an image are literally pixel-locked.
2. Repeated edits should drift less
Generating the first image has rarely been the hardest part of AI image creation.
Revision seven is where things tend to fall apart.
You change a headline, then adjust the background, then move an object, and by the fourth or fifth instruction the image no longer quite resembles the version you started with.
Images 2.5 is designed to retain previous decisions more reliably during multi-turn editing. Earlier changes are more likely to survive while subsequent instructions are applied.
That improvement could matter more to working designers and marketers than another jump in photorealism.
A usable creative tool needs memory across revisions. Otherwise, every edit becomes another roll of the dice.
3. You can draw what you mean with ChatGPT Sketch
One of the genuinely new additions is Sketch.
Instead of describing the exact positioning of every element in a paragraph, you can make a rough drawing and ask ChatGPT to turn it into a finished image.
OpenAI gives examples such as sketching a room layout or the outline of an outfit. You can then describe the visual style and other details you want the final image to contain.
OpenAI’s current instructions document Sketch on the ChatGPT mobile app:
type @ in the message box → select Sketch → draw the basic idea → confirm the drawing → add your written instructions → generate the image.
The feature is more important than it initially sounds.
Text is not always a good interface for visual positioning. Writing “put the bottle slightly below the headline but aligned with the right-hand edge of the product card” is far less natural than drawing approximately where the bottle should go.
The Verge tested Sketch with a simple cat doodle and reported that ChatGPT was able to turn it into a more detailed image and then perform subsequent visual edits.
4. Comments make image editing more visual
Images 2.5 also lets users add comments directly to areas of an image rather than describing every change only through the main conversation.
That turns the workflow into something closer to reviewing a design with another person.
You can point to the area you care about, explain the change and leave the rest of the image alone.
OpenAI’s Help Center notes that selections are not always perfectly precise and edits can occasionally extend outside the selected region, so the editor should still be treated as generative rather than as a Photoshop-style pixel editor.
That limitation is worth knowing before using it for finished catalog photography or tightly controlled brand assets.
ChatGPT’s image editing interface
The first image shows OpenAI’s selection-based editing workflow. The second shows the Images interface where users can start image creation and explore visual options.
5. Templates reduce the blank-prompt problem
OpenAI has also introduced Templates.
Instead of beginning every request with an empty prompt box, users can start with formats such as posters, logos or merchandise and customize them with text, imagery and design preferences.
This is not the most technically impressive part of Images 2.5, but it could be one of the most used.
Most people do not want to become expert prompt engineers just to create a flyer.
Templates shift some of that work into the interface. ChatGPT can also ask follow-up questions when it needs additional information about the intended visual.
One current limitation is that Templates are not yet available in ChatGPT Work mode, according to OpenAI’s Help Center.
6. Prompts can now travel with the image
ChatGPT now lets users share the prompt associated with a generated image.
Someone receiving that prompt can reuse the underlying idea while substituting their own photos or details.
That could turn image prompts into something closer to reusable creative templates.
For teams, it also provides a much easier way to share a successful concept than sending screenshots and attempting to reconstruct the instructions manually.
Where Images 2.5 should be noticeably more useful
The upgrade seems particularly relevant to workflows where iteration matters.
| Use case | Why Images 2.5 helps |
|---|---|
| Product photography | Change backgrounds or details while preserving the product |
| Social media creative | Produce variations without rebuilding the concept |
| Advertising | Refine copy, objects and compositions over several revisions |
| Brand mockups | Maintain visual identity across changes |
| Posters and flyers | Improved layout handling plus templates |
| Photo transformations | Better preservation of recognizable subjects |
| Presentations | More dependable structured visuals and transparent backgrounds |
| Rapid prototyping | Sketch an idea before writing a detailed creative brief |
OpenAI specifically positions the new system for creator content, product experiences, visual search, prototyping and production creative workflows.
GPT-Image-2.5 Flare vs Sunburst for API users
Developers get two versions of the new model.
GPT-Image-2.5 Flare is OpenAI’s default recommendation for most applications. It prioritizes speed while retaining the new model’s improvements in image quality and editing.
GPT-Image-2.5 Sunburst is aimed at more demanding work where editing precision matters more than generation time. OpenAI describes use cases such as polished product photography and production-ready campaign creative.
| API model | Best suited for |
|---|---|
| GPT-Image-2.5 Flare | High-volume generation, social content, prototypes, everyday product experiences |
| GPT-Image-2.5 Sunburst | Detailed editing, campaign assets, polished commercial imagery |
Both currently list text input at $5 per million tokens, image input at $8 per million image tokens, and image output at $30 per million tokens. Cached inputs are priced lower. Actual cost per generated image depends on the inputs, output and quality settings used.
How to use ChatGPT Images 2.5
There is no separate ChatGPT Images 2.5 application you need to install.
Open ChatGPT and either describe the image you want in a conversation or open Images from the interface. Existing images can be uploaded and edited using natural-language instructions. ChatGPT also supports specifying details such as text and aspect ratio.
For better results, a useful prompt structure is:
| Prompt element | Example |
|---|---|
| Subject | A premium stainless-steel coffee maker |
| Setting | On a walnut kitchen counter |
| Composition | Product on the right, clear space on the left |
| Lighting | Soft natural morning light |
| Style | High-end editorial product photography |
| Text | “Better Coffee. Less Effort.” |
| Format | 16:9 website hero image |
| Constraint | Keep the logo and product proportions unchanged |
Images 2.5 should reduce the need to rewrite the entire prompt when something is wrong. Once you have the composition you want, subsequent requests can focus on the particular element that needs changing.
Is ChatGPT Images 2.5 actually a big upgrade?
It depends on what you use image generation for.
If your workflow consists of entering a prompt, choosing one image and moving on, the difference may feel incremental: sharper output and shorter waiting time.
If you routinely edit, revise, reuse reference photos or create several versions of the same asset, Images 2.5 addresses much more consequential problems.
The move toward sketches, comments and templates is also revealing. OpenAI is gradually moving image generation away from a prompt box where users describe an image and toward a more interactive creative workspace.
That is probably the more important story behind Images 2.5.
The model is getting better, but the interface between the person and the model is changing too.
ChatGPT Images 2.5 FAQs
Is ChatGPT Images 2.5 free?
ChatGPT Images is available across ChatGPT tiers. Usage limits can still vary by plan, and OpenAI says existing image-generation limits remain unchanged with the 2.5 launch.
When was ChatGPT Images 2.5 released?
OpenAI announced ChatGPT Images 2.5 on September 8, 2026.
Is ChatGPT Images 2.5 available on mobile?
Yes. ChatGPT Images is available on web, iOS and Android. Sketch is currently documented specifically for the ChatGPT mobile app.
What is ChatGPT Sketch?
Sketch lets you draw a rough visual directly in ChatGPT and then use that drawing as a reference for image generation.
What is the difference between ChatGPT Images 2.5 and GPT-Image-2.5?
ChatGPT Images 2.5 is the image-generation experience inside ChatGPT. Developers accessing the technology through OpenAI’s API can choose GPT-Image-2.5 Flare or GPT-Image-2.5 Sunburst.
Can ChatGPT Images 2.5 edit existing photos?
Yes. You can upload an existing image and request changes. The new model is specifically designed to improve reference fidelity, targeted editing and consistency across repeated edits.
Is DALL-E still available inside ChatGPT?
OpenAI retired its official DALL-E GPT on August 30, 2026 and now directs users toward ChatGPT Images for image creation and editing. User-created GPTs with image-generation capabilities are not affected in the same way.
Final thoughts
ChatGPT Images 2.5 is not interesting simply because the pictures look a little better.
Its more useful improvements happen between the first image and the finished one.
Faster generation reduces waiting. Better subject preservation makes reference images more dependable. Multi-turn consistency reduces the chances of losing earlier work. Sketch and comments give users ways to communicate visually when text alone is awkward.
That combination pushes ChatGPT closer to being a practical image-editing workspace rather than just another text-to-image generator.
For people producing marketing creative, product imagery, social assets, mockups or visual concepts regularly, that distinction matters.
Snehil Prakash is a growth leader in saas and it industry. He has lead multi million dollar businesses in this space and currently the CEO of Techlithic solutions LLC that owns brands like howtobuysaas, sendwo and work as venture partner for businesses to raise captial or going to the IPO way. Reachout to him to learn more.
