OpenAI speeds up Images 2.6 by 50% and adds GIF animation to the API

Image generators often fail not on the first frame, but on the third edit. Ask to replace the background, and the product changes shape. Fix a character's hand, and the clothing shifts. Add another object, and the entire composition has to be rebuilt. Images 2.6 targets this familiar problem directly: it enables faster sequences of edits without losing important details along the way.

OpenAI says the new version generates images 50% faster than its predecessor. The API offers two models: GPT-Image-2.6 Flare for speed and GPT-Image-2.6 Sunburst for higher quality and more consistent results across iterations. At the low-quality setting, Flare can produce an image in about six seconds.

What has changed in the workflow

Consider an online store that needs images of the same backpack for its website, ads, and social media. Previously, every edit could accidentally alter the clasp, the fabric color, or the logo. Developers can now select the required area using image region markup and request a change only to the background or lighting. Sunburst is designed for cases where a product's or character's appearance must remain intact across several such rounds.

The API also introduces a complexity parameter. It determines how much effort the model spends constructing a scene. For a simple background replacement, OpenAI recommends low complexity so the response arrives faster. If a frame contains many objects, poses, and constraints, complexity can be raised to medium or high. Meanwhile, quality controls image detail, while complexity controls compositional difficulty. The two settings can be adjusted independently.

Another practical addition is SVG export. It is useful for illustrations, interface elements, and print materials that must scale without losing sharpness. But SVG does not turn every result into a flawless design file: geometry, layer structure, and the file's suitability for further editing still need to be checked.

A GIF from a single prompt

Images 2.6 can create a looping GIF animation. For example, a brand can describe six consecutive states of a mascot: the character notices an event, reacts, moves, and returns to the starting pose. The model prepares frames with a more consistent character appearance, which can then be used for a reaction, an advertising creative, or a rough storyboard.

This is not a replacement for a full-fledged video generator. GIF is suitable for short, repeating motion that does not require sound, complex camera work, or long-form storytelling. For an animation prototype or a series of social media creatives, however, the format is cheaper and easier to integrate into an automated pipeline.

What developers and product teams should do

Do not move the entire production workflow to maximum settings immediately. Start by assembling a small test set: a precise product edit, character consistency across several scenes, a complex poster, and a short GIF. Compare speed, the number of failed iterations, and the amount of manual rework required with Flare and Sunburst.

For high-volume drafts, it makes sense to start with Flare, low complexity, and a quality level appropriate to the task. Sunburst is better reserved for final assets where an error in a product's shape or a brand detail costs more than the extra wait. Brand review, moderation, and artifact checks are still required before publication: more consistent generation reduces errors but does not eliminate them.