Images 2.6 adds GIFs and finer control over revisions

OpenAI launched Images 2.6 on November 4, announcing global availability in ChatGPT and Codex across subscription tiers and two API models, GPT-Image-2.6 Flare and Sunburst. The release adds GIF creation and brings image markup, SVG export and a complexity control to the API. OpenAI also claims faster generation and improved consistency through edits; NEXTECH has not independently measured those improvements.

Quality and planning answer different questions

The announcement separates image quality from complexity: one concerns visual fidelity, the other planning for a scene. That distinction is useful when defining a job. Replacing a background while keeping an approved product illustration stable is different from inventing a composition with several interacting subjects. Our reading is that creative teams should specify what must survive an edit before choosing how much variation they want. A visually impressive replacement can still fail that brief.

NEXTECH analysis: count the accepted revision

We would measure the time from a written brief to an accepted revision. A fast first image offers limited savings if the next change alters the product silhouette, moves an approved label or requires rebuilding the layout. A representative evaluation could begin with a fixed reference and request several small changes in sequence, recording both the requested edits and any unintended ones. This is our proposed test, not a result for Images 2.6. It would reveal whether consistency persists across the ordinary corrections that publication work requires.

Motion and export add their own inspection work

The additional formats also change what acceptance means. For a short loop, we would inspect the transition back to the first frame and check whether important details remain stable throughout playback. For an SVG, we would inspect the exported document at the intended size and try the edits a designer actually needs. The existence of an export option cannot establish that every resulting file has convenient structure, correct text or the same appearance in every destination. Those are questions for the delivered asset.

What still needs an independent comparison

Our judgment is that control over revision deserves as much attention as an attractive generation sample. A small comparison using a team's real briefs would be more informative than unrelated showcase images: preserve the same reference, record the settings and count how many outputs meet the original acceptance criteria. We have not performed that comparison, inspected exported files or assessed animation continuity. The announcement establishes new creative options; the cost and reliability of reaching a publishable result remain to be measured.