Quick Answer
ChatGPT Images 2.5 adds Sketch, templates, and more targeted editing while OpenAI says the update can cut generation latency by up to 50% versus Images 2.0. The September 8 rollout reaches ChatGPT, ChatGPT Work, and Codex users across all tiers. Use Sketch for visual direction, but review generated images carefully, especially where realism could mislead.
Key Takeaways
- ChatGPT Images 2.5 began rolling out on September 8, 2026, across ChatGPT, ChatGPT Work, and Codex.
- OpenAI says the updated image system delivers up to 50% lower generation latency than Images 2.0.
- The new Sketch feature combines a drawn reference with written instructions.
- Templates support formats such as flyers and product photos, but templates are not yet available in ChatGPT Work.
- OpenAI says the system blocks sensitive deepfakes, although users should still assess realistic images carefully.
What is ChatGPT Images 2.5?
ChatGPT Images 2.5 is OpenAI’s updated image-generation and editing system for ChatGPT, ChatGPT Work, and Codex. OpenAI announced the release on September 8, 2026, describing sharper details, more precise edits, stronger preservation of reference photos, and generation latency that is up to 50% lower than Images 2.0. OpenAI’s product announcement describes the update as a consumer-facing improvement to image creation rather than a separate ChatGPT app.
The lower-latency claim matters because image generation often involves several attempts before a user gets the composition, lighting, or object placement they need. Faster results can make that iterative process less disruptive, particularly when a user is making small corrections to an existing image. OpenAI’s figure is a company claim, however, and the announcement does not provide a single fixed generation time for every prompt, device, or image request.
ChatGPT Images 2.5 also arrives during a broader period of AI product releases, including OpenAI’s recent computer-task model update. The practical distinction is that Images 2.5 focuses on visual creation and editing inside ChatGPT, while users should evaluate each feature based on the task they need completed.
Who can use ChatGPT Images 2.5?
ChatGPT Images 2.5 is rolling out to ChatGPT, ChatGPT Work, and Codex users on desktop, mobile, and web across all tiers. OpenAI says the rollout began on September 8, 2026, which means availability can still vary while the update reaches individual accounts and platforms. Users who do not see the new controls immediately should check the current ChatGPT interface and allow time for the rollout to complete.
ChatGPT Images 2.5 does not change the existing image-generation limits. OpenAI’s release notes also state that templates are not yet available in ChatGPT Work, so access to the broader update does not mean every account receives every tool at the same time. OpenAI’s ChatGPT release notes identify the availability difference, which is important for workplace users who expect feature parity with consumer ChatGPT accounts.
ChatGPT Images 2.5 is designed for users who need to create images from prompts, revise a generated image, or provide a visual reference. The most sensible approach is to confirm which controls appear in your own account before planning a workflow around templates or Sketch, especially if your work depends on a shared team environment.
How does the ChatGPT Images 2.5 Sketch tool work?
ChatGPT Images 2.5 Sketch lets users combine a simple drawing with written instructions to create a finished image. Users type “@Sketch” in ChatGPT, draw a visual reference, and then add a prompt describing the desired result. The feature gives the model a rough visual guide for layout, shape, positioning, or composition while the written request supplies the details.
ChatGPT Images 2.5 Sketch is particularly useful when words alone do not clearly describe the image arrangement you want. A rough drawing can show where an object belongs or how a scene should be framed, while the prompt can describe materials, colors, style, and other attributes. OpenAI product lead Adele Li told Axios that Sketch and templates are meant to make users more active participants in the creative process. Axios’s interview and hands-on report provides that product-lead perspective.
ChatGPT Images 2.5 Sketch does not guarantee that every generated result will follow a rough drawing exactly. Image models interpret both visual and written inputs, so users should inspect object placement and important details before using an output publicly or professionally. The practical response is to make the drawing clear, state the most important requirements in the prompt, and request a targeted revision when the first result misses a key detail.
- Type “@Sketch” in the ChatGPT message box.
- Draw a visual reference that shows the basic layout or object placement.
- Add written instructions describing the intended finished image.
- Review the generated image for composition, details, and unintended changes.
- Request an edit when the image does not preserve the required visual elements.
Which editing tools does ChatGPT Images 2.5 add?
ChatGPT Images 2.5 adds templates, image comments, and prompt sharing alongside the Sketch feature. OpenAI says templates support formats including flyers and product photos, while image comments allow more targeted edits to parts of an image. Prompt sharing gives users a way to pass a visual request to another person without requiring them to rebuild the instruction from scratch.
ChatGPT Images 2.5 templates can make common visual formats easier to start, but templates are not yet available in ChatGPT Work. That limitation matters because a workplace account may have access to image generation without access to the same preset formats available elsewhere. Users should begin with a template when it is available, then use image comments or follow-up instructions to correct specific parts of the result.
| ChatGPT Images 2.5 tool | What it does | Main limitation |
|---|---|---|
| Sketch | Combines a drawing and written prompt to guide image creation. | Generated images can still interpret a rough drawing imperfectly. |
| Templates | Provides starting formats such as flyers and product photos. | Templates are not yet available in ChatGPT Work. |
| Image comments | Supports targeted feedback on specific image areas. | Users still need to review whether the requested edit was applied accurately. |
| Prompt sharing | Lets users share an image-generation instruction. | Shared prompts can produce different outputs when details change. |
ChatGPT Images 2.5 expands the ways users can guide an image model, but it does not remove the need for judgment. A flyer, product photo, or shared prompt can be a useful starting point, while the final image still requires a review for factual accuracy, brand requirements, and sensitive visual details.
How do GPT-Image-2.5 Flare and Sunburst differ?
GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst are two API models that OpenAI introduced with the Images 2.5 release. OpenAI positions Flare for faster, high-quality image generation and Sunburst for more precise, detailed work that takes longer to generate. The split gives developers a choice between quicker output and a slower model intended for requests where fine details matter more.
GPT-Image-2.5 Flare is the better fit when an application values speed and needs a strong general-purpose image result. GPT-Image-2.5 Sunburst is the better fit when a workflow needs more detailed output and can accept a longer generation time. OpenAI does not provide a universal generation-time figure for either API model in the release details, so developers should avoid assuming a fixed speed difference for every request.
ChatGPT users do not need to select Flare or Sunburst as part of the consumer interface described in the announcement. The API distinction matters primarily to developers building image features into their own products, much as organizations weigh model options in tools such as multi-model coding services. The practical decision is to choose the faster option when responsiveness is the priority and the more detailed option when precision justifies a longer wait.
What are the safety limits of ChatGPT Images 2.5?
ChatGPT Images 2.5 can create more convincing images, which increases the risk that realistic outputs could be used as political, sexual, or otherwise sensitive deepfakes. OpenAI’s system card identifies that risk and says the company uses prompt-layer and image-layer safeguards to block such content. OpenAI’s Images 2.5 system card explains that added realism creates a higher potential for harmful misuse even when safeguards are in place.
ChatGPT Images 2.5 safeguards reduce risk, but safeguards do not make every realistic image trustworthy or eliminate misuse outside the service. A convincing image can affect a person’s reputation, influence public understanding, or create confusion when its origin is unclear. Users should not present AI-generated visuals as evidence of real events, real people, or real products without clear verification and appropriate disclosure.
ChatGPT Images 2.5 also requires care when users upload reference photos. Reference preservation can be useful for legitimate creative work, but users should avoid sharing sensitive personal images, confidential workplace materials, or images of other people without permission. The most sensible approach is to use only references you have a right to upload and to stop before creating or distributing a realistic image that could harm, impersonate, or mislead another person.
What should users do first with ChatGPT Images 2.5?
ChatGPT Images 2.5 users should begin with a low-risk creative request, then test Sketch or targeted edits on an image that does not contain sensitive personal or business information. OpenAI says people create more than 3 billion images each week across ChatGPT Images and GPT-Image API models, which shows that image creation is already a large-scale part of the company’s AI ecosystem. That scale does not change the need to review each output before sharing it.
ChatGPT Images 2.5 works best when a prompt separates the essential visual requirements from optional style preferences. Users can use Sketch to establish a layout, then state the subject, setting, visual style, and required details in the accompanying text. The update promises better reference-photo preservation, but a careful review remains necessary because an image can include errors even when it generally follows the request.
ChatGPT Images 2.5 users should also treat shared prompts as reusable starting points rather than finished instructions. Prompt sharing can help teams establish a common format, while individual outputs can still differ as users alter references or add new details. Similar concerns apply to personal AI tools that can act on user information, including AI agents that make purchases, where convenience does not remove the need to confirm the final action.
FAQ
Is ChatGPT Images 2.5 available to free users?
ChatGPT Images 2.5 is rolling out across all tiers of ChatGPT, ChatGPT Work, and Codex. OpenAI says the rollout began on September 8, 2026, so individual accounts may receive the update at different times.
How do I open Sketch in ChatGPT on mobile?
ChatGPT Images 2.5 Sketch opens on mobile when you type “@” in the message box and select Sketch. Add a drawing and written instructions to guide the resulting image.
Does ChatGPT Images 2.5 make images faster?
ChatGPT Images 2.5 can deliver up to 50% lower generation latency than Images 2.0, according to OpenAI. Actual timing can vary because the company does not publish one fixed generation time for every image request.
Are ChatGPT Images 2.5 templates available in ChatGPT Work?
ChatGPT Images 2.5 templates are not yet available in ChatGPT Work. ChatGPT Work users can still receive the broader Images 2.5 rollout, but they should not expect every consumer-facing feature to appear in their account.
Can ChatGPT Images 2.5 create harmful deepfakes?
ChatGPT Images 2.5 has a higher deepfake risk because improved realism can make sensitive images more convincing. OpenAI says prompt-layer and image-layer safeguards block political, sexual, and other sensitive deepfake requests, but users should not treat generated images as evidence of real events.