TL;DR

OpenAI released Images 2.6 on 4 November across ChatGPT, Codex and its API, adding animated GIFs and promising generation 50% faster than before. The company gives no UK-specific detail, but it describes the release as global and across all subscription tiers. For UK marketing and design teams, the practical change is finer control over edits and brand consistency, and all the performance claims here are OpenAI’s own.

What is new

Motion is the headline. People can now turn a sketch, mascot or product shot into a looping animation. OpenAI pitches this for campaigns, social media reactions and storyboards tested before video production. You can see the launch clip on our Videos page or above.

Developers get two models. GPT-Image-2.6 Flare is the fast one, which OpenAI says can return an image in six seconds at best with quality set to low, and it is also said to improve layouts on posters and product imagery. Sunburst is pitched at higher quality and better consistency between edits at medium or high quality settings.

Image markup moves from ChatGPT to the API, letting teams mark exactly what should change, and SVG export makes output usable on websites, in apps and in print. Separately from image quality, developers can now set how much the model plans through a complexity setting; OpenAI recommends low complexity for most edits.

Scale is the other claim: 4 billion images a week across its products, or over half a billion daily. The figures come only from OpenAI.

Where this leaves UK users

Availability is simple, since the announcement lists every subscription tier and all its apps. OpenAI’s system card reports its own safety tests on 3,077 adversarial requests. Unsafe images getting past moderation fell for Sunburst, from 1.92% to 1.72%, and rose for Flare, from 1.49% to 2.14%, against the previous versions. OpenAI says neither change was statistically significant, and that the tests do not measure production use. Before moderation, Flare’s share of safe outputs fell significantly, from 78.36% to 75.04%. The card lists training data only in broad categories and does not address rights in the output, so UK firms should not read it as settling those issues. It also says images carry C2PA metadata and SynthID watermarks.

This is the latest step since Images 2.0 in April. The pace of these updates means that a workflow signed off in the spring may already be out of date.

Looking Forward

Early customer examples aside, independent testing will show whether the speed and consistency claims hold in practice. In our view, UK brand teams should try the GIF and consistency features on low-risk assets first, and check their approval processes. An animated mascot is quick to make, which means more material for marketing and legal teams to review, under existing advertising and intellectual property rules, before it goes live.