The volume of synthetic imagery circulating online has grown steadily as tools for generating and refining pictures have become more accessible.
Across major platforms, users now produce images at a scale measured in the billions each week, with daily totals exceeding half a billion. OpenAI, the creator of ChatGPT, has released an update to its image generation system.
The release, called 'Images 2.6,' is available in ChatGPT, Codex, and through the API.
It reduces generation time by about half relative to the prior version.
In the API, two variants are offered. GPT-Image-2.6 Flare prioritizes speed and can return an image in as little as six seconds when quality is set to low. It also improves layout consistency for items such as posters and product shots.
GPT-Image-2.6 Sunburst is intended for higher detail and better preservation of specific elements across successive edits when medium or high quality settings are used.
A notable addition is support for animated GIFs.
Users can generate a looping sequence from a text description or an existing still image, then adjust playback speed and modify individual frames.
Examples shown in the announcement include short character animations, such as a stylized office worker reacting to a changing stock chart, and simple product or mascot loops. The same capability extends to the API, where developers can specify frame count and duration.
Reference images can now be stored for repeated use.
Once a photograph of a person, pet, or character is saved, it can be called in later prompts without re-uploading. This applies within ChatGPT and is positioned as a way to maintain visual continuity across multiple generations.
For developers working through the API, several controls were added or extended.
Image markup, previously limited to the chat interface, allows a transparent overlay to indicate the exact region of an existing image that should change. SVG export produces vector output suitable for scaling in web or print contexts.
A complexity parameter lets callers indicate how much planning the model should apply before rendering, with lower settings recommended for straightforward edits and higher ones for scenes involving multiple constraints.
Pricing for the new models remains at the same per-token rates as the previous generation.
The update is available globally across all ChatGPT and Codex subscription tiers on web, desktop, and mobile. Early users of the API version, including teams at OpenArt, Figma Weave, and Picsart, described faster iteration and improved consistency when refining characters or product visuals, though these accounts come from the announcement itself.
In practical terms, the changes center on reducing the number of separate steps required to move from an initial idea to a usable still or short animation, particularly when the same subject must appear repeatedly or when precise local edits are needed.
