Images 2.6 splits speed and fidelity into two API choices
OpenAI has released Images 2.6 across ChatGPT, Codex and the API. The developer offering is divided into GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst. Flare targets rapid iteration and can produce a low-quality image in as little as six seconds, while Sunburst is positioned for higher detail and more consistent edits at medium or high quality.
OpenAI says the generation system is 50% faster than its predecessor. That headline will be most valuable in workflows where people create many candidates before selecting one, such as product concepts, storyboards and campaign variations. Teams should benchmark complete review cycles, because a fast first render does not save time if brand details or text require repeated correction.
GIF generation brings motion into the image workflow
The release can generate looping GIFs from a written idea or an uploaded sketch. OpenAI highlights mascots, reactions, memes and animated storyboards as likely uses. This fills a practical gap between a static image tool and a full video system: creators can test movement and personality without building a timeline or paying the compute cost of a longer video generation.
Loop quality remains a demanding test. A convincing result needs stable characters, controlled camera position and a final frame that returns naturally to the first. Production teams should inspect flicker, accidental text, brand drift and accessibility implications. GIFs have no native audio and can become large, so delivery format, reduced-motion alternatives and compression still belong in the publishing workflow.
Markup and saved references reduce editing friction
Image markup, already used inside ChatGPT, is now exposed through the API so applications can identify the part of an image that should change. Saved reference images in ChatGPT allow a user to name an asset and reuse it in later conversations. Together, these features make iterative art direction more precise than repeatedly describing a location or re-uploading the same source.
The convenience introduces data-management questions. A saved person, product or logo becomes a reusable identity asset and needs clear ownership, retention and deletion controls. Applications should record which source image and edit instructions produced a final asset. That provenance helps teams reproduce approved work, respond to rights questions and avoid silently carrying an outdated brand reference into a new campaign.
SVG export and complexity controls target professional pipelines
Images 2.6 can export SVG for designs that need to move into websites, applications or print. A new complexity parameter controls how much planning the model applies, separately from visual quality. OpenAI recommends low complexity for most editing tasks and higher settings for scenes with many elements or constraints. This gives developers a more useful latency-versus-control decision than one quality slider.
Generated SVG should still be treated as untrusted input before it reaches a website or design system. Teams need to sanitise scripts, external references and embedded data, then inspect geometry and text at multiple sizes. For print, colour profiles, bleed and font handling remain conventional production concerns. Model output can accelerate drafting without replacing preflight and brand review.
Availability is broad, but evaluation should stay specific
Images 2.6 is available globally in ChatGPT and Codex across subscription tiers on desktop, mobile and web. Flare and Sunburst are available through the API. The broad rollout means organisations can compare the conversational and programmatic experiences immediately, including how well instructions survive when assets move from experimentation into a repeatable application workflow.
A useful evaluation set should reflect the organisation’s real work: recurring characters, products with exact geometry, layouts with readable text, local cultural requirements and edits that must preserve everything outside a marked region. Measure acceptance rate, edit count, latency and cost together. The new controls matter when they reduce the distance between an idea and an approved asset, not merely when they create more images faster.
Creative teams need a production contract for generated assets
An organisation adopting Images 2.6 should define which outputs can be published directly and which require design, legal or product review. Brand characters, customer likenesses and regulated claims need stronger controls than an internal concept sketch. File naming, version status and approval metadata should travel with the asset so a fast generation workflow does not create ambiguity about what is cleared for use.
Prompt and reference libraries can become valuable intellectual property. They capture the instructions, examples and constraints that make a model reproduce a distinctive visual language. Access should be limited like other design assets, and teams should avoid storing rights-restricted material in a shared reference pool. A reusable workflow is only an advantage when ownership and permissible reuse are clear to everyone who can invoke it.
Developers should also test API failure behaviour. A partially completed edit, safety refusal or model upgrade can break a batch pipeline differently from an interactive chat. Stable request records, output validation and a human review queue make creative automation recoverable. Images 2.6 supplies more capable building blocks; production reliability still comes from the application surrounding them and the people accountable for the final image.