ChatGPT Images 2.6: More consistent GIFs and editing
OpenAI introduced ChatGPT Images 2.6 with faster generation, GIFs, and greater consistency across edits. We look at what's changing, how saved references work, and which tools remain on the integration side for developers.
OpenAI introduced ChatGPT Images 2.6 on November 4, 2026, with improvements for generating images and preserving their details during edits. The update adds GIFs and saved references that you can reuse in other conversations, according to the official announcement.
The change reaches ChatGPT and Codex, while developers get two models: GPT Image 2.6 Flare and Sunburst. The features have different access and performance conditions; just because the model is available doesn't mean all its new features show up at the same time in every account.
What changes when editing an image
OpenAI says Images 2.6 better maintains visual continuity between changes. The goal is to preserve the characteristics of a character or product while modifying another part of the composition.
One example use would be replacing the background of an illustration without changing the character's clothing. The announced improvement points to that continuity. It does not establish that every edit will preserve all details exactly.
The official image prompting guide keeps one specific recommendation: state what should change and what should stay the same. It also suggests refining one aspect at a time and reviewing the result.
That helps put the scope of the announcement in perspective. Consistency describes a model capability; the fidelity of a particular piece still depends on the result it delivers.
The documentation includes examples of localized changes, such as removing an object or replacing elements within a photograph. In those cases, the instruction identifies both the modification and the elements that must be preserved.
GIFs from images and descriptions
GIF creation lets you turn a description or an image into an animation that plays on a loop. Among the announced uses are animating a character and giving movement to a product image.
The ChatGPT release notes indicate that you can adjust the playback speed and share the GIF. They also clarify that the feature is rolling out gradually.
This introduces a difference compared with generating a static image: the result needs continuity between frames and repetition that makes sense. A short animation can show a gesture or movement without becoming a full video production.
That capability shouldn't be extended to any audiovisual output. A looping GIF is a specific format, with a narrower scope than a long video sequence.
Saved references across conversations
Another addition is saving reference images to use later, without uploading the same file again. The release notes describe reusing them through a mention by name.
The feature changes how you retrieve your source material. If a reference is already saved, you can invoke it in another conversation for a new creation.
That can be relevant for anyone working with a recurring character. Starting from the same reference makes it easier to communicate which appearance you want to keep, though it isn't a guarantee of perfect visual identity in every output.
There's also a difference between saving the reference and editing a previous image. The reference provides material for a new request; editing starts from a specific image you want to modify.
Flare and Sunburst have different priorities
To integrate generation into an application, OpenAI offers two variants. The models and instructions documentation presents Flare as the option focused on speed and Sunburst as the alternative for greater detail and precision when editing.
| Model | Approach described by OpenAI |
|---|---|
| GPT Image 2.6 Flare | Generation and editing with a priority on speed |
| GPT Image 2.6 Sunburst | Visual work with higher demands for detail and precision |
The names correspond to models available through the API, the connection that makes it possible to incorporate these capabilities into another system. They should not automatically be interpreted as two buttons that every ChatGPT account will display the same way.
The official changelog dates its release on November 4 and confirms new generation and editing capabilities.
What the six-second figure means
OpenAI announces a generation that is 50% faster and notes that Flare can produce images in as little as six seconds when low quality is used. These are the manufacturer's claims, published in its Images 2.6 presentation.
The condition matters: six seconds is not a guaranteed time for any request in ChatGPT. The figure corresponds to a specific variant and setting.
The generation guide also separates two controls that can be confused. Quality regulates the image's detail; complexity regulates how much the model plans before producing it.
According to the generation documentation, higher complexity can help with demanding compositions, but it can also increase wait times and introduce unwanted changes. More planning does not ensure a more faithful edit in every case.
Additional tools for integrations
The update brings guided editing via marks on an image to the API. This makes it possible to indicate where a change should be made, instead of relying solely on a written description.
It also adds SVG output, a graphics format that can be scaled without the loss of sharpness typical of an image made up of pixels. The changelog confirms this capability for the new models.
These tools expand integration possibilities. Their presence in the API does not prove that an identical option exists within every ChatGPT interface.
Availability with feature-specific conditions
The OpenAI help center states that ChatGPT Images is available on all subscription tiers. The reasoning-based image mode is listed for Plus, Pro, and Business; Enterprise and Edu appear as coming soon.
The same documentation explains that you can create images from a conversation or upload an existing one to request changes. It also supports instructions to incorporate text and produce transparent backgrounds.
The scope of Images 2.6 lies in combining more consistent editing with new ways to reuse and animate visual material. The availability of each feature should be read separately: general access to images, the rollout of GIFs, and the integration tools each have their own conditions.
Terms of this note
- GIF: image format that shows a short animation, usually looped.
- Saved reference: an image stored so it can be reused in later requests without uploading it again.
- API: a connection that lets one application use features from another service.
- Flare and Sunburst: the names of the two GPT Image 2.6 variants for developers.
- Complexity: a setting that controls how much the model plans before generating or editing.
- SVG: a vector graphics format that lets you enlarge its elements without losing sharpness.
- Reasoning images: a mode in which the system devotes extra processing to preparing the visual creation.
References
Information reviewed on November 9, 2026. The launch date is November 4. Performance figures are OpenAI's claims, not our own measurements.