The unreleased interface promises higher-quality results and smarter tools, while OpenAI's public documentation still lists Images 2.0.
By Ryan Merket · Published
Primary source: X - @chetaslua
Why it matters #
OpenAI is preparing to make image generation faster and more iterative just days after retiring its standalone DALL-E GPT, tightening ChatGPT's grip on visual creation.
OpenAI appears to be preparing another upgrade to ChatGPT's image generator, according to an unreleased product modal surfaced inside Codex on September 3rd.
Chetaslua (@chetaslua) posted a screenshot on X showing a hidden modal with the headline "Image creation got a major upgrade." The interface promises "higher-quality results, faster generation, and smarter creative tools" and includes a "Try it" button.
Chetaslua identified the update as "ImageGen 2.5" and said the modal was already built into the product but remained behind a disabled feature gate. The 2.5 name does not appear in the visible interface, so the screenshot alone establishes the presence of an unreleased image-generation upgrade, rather than its final public branding.
The researcher also said the modal's hero asset was uploaded on September 1st. That timing, combined with the completed onboarding interface and call-to-action button, points to a release package that has moved beyond an early design mockup. It does not establish when OpenAI plans to activate the feature or which ChatGPT plans would receive it first.
OpenAI still publicly lists Images 2.0
OpenAI's public materials had not caught up with the hidden interface as of September 3rd. The company's ChatGPT release notes list a healthcare data update under September 1st and sticker creation under August 31st, with no entry for ImageGen 2.5 or another core image model.
A separate OpenAI help page for image creation, updated within the previous day, still identifies ChatGPT Images 2.0 as the model available across all subscription tiers. That documentation says users can create and edit images on the web, iOS and Android, while paid customers can access an image mode that spends additional reasoning time planning an output.
OpenAI released ChatGPT Images 2.0 on April 21st. OpenAI described that release as improving world knowledge, instruction following and the generation of dense text and detailed scenes. Its accompanying thinking mode could search the web, generate multiple images from one prompt and use OpenAI's reasoning stack to expand a basic request before producing the final image.
The leaked modal uses broader consumer language. It promises better quality, speed and creative controls without naming a model architecture, benchmark, resolution, pricing structure or API endpoint. Those omissions leave open whether ImageGen 2.5 represents a new underlying model, an updated inference stack, a product-layer bundle of editing tools, or some combination of the three.
Speed is the clearest product promise
Faster generation is the most concrete claim in the unreleased copy. OpenAI's current help documentation warns that image creation can take several minutes depending on the complexity of a request. Reducing that delay would make image generation easier to use iteratively, especially for workflows that require repeated changes to text, composition or aspect ratio.
The reference to "smarter creative tools" also suggests OpenAI is packaging the model update around the editing experience rather than treating image generation as a single prompt-and-output feature. ChatGPT already supports regional selection, direct conversational edits, transparent backgrounds and aspect-ratio changes. The modal could introduce additional controls or improve how reliably those existing tools follow instructions.
The timing follows OpenAI's August 30th retirement of the official DALL-E GPT from ChatGPT. OpenAI directed users toward ChatGPT Images instead, consolidating image creation and editing under the newer product. A 2.5 update would give OpenAI a cleaner capability jump to promote immediately after completing that transition.
The hidden modal shows that the next image release already has customer-facing language and artwork inside OpenAI's product. Its final name, rollout schedule and technical scope remain outside the visible interface. Until OpenAI activates the gate or updates its documentation, "ImageGen 2.5" remains an externally reported internal label attached to a clearly prepared upgrade.