OpenAI’s ChatGPT Images 2
- OpenAI launched a new image model, ChatGPT Images 2, with native reasoning, 2K resolution and multi‑image consistency. - Coverage says the model “thinks” before drawing and is already facing stress tests from viral misuse experiments. - Early real‑world probing is forcing immediate governance questions about moderation, cohort rollouts and abuse detection (cnet.com).
OpenAI rolled out ChatGPT Images 2.0 on April 21, adding a new image model that can reason through prompts before it draws. (openai.com) (thenewstack.io) The new model is available in ChatGPT across all tiers, while the “images with thinking” mode is limited to Plus, Pro, and Business for now; OpenAI’s help center says Enterprise and Edu access is still coming. (help.openai.com) OpenAI and early coverage say the upgrade adds 2K output, stronger text rendering inside images, and multi-image consistency, which is the ability to keep the same character or design details across several pictures. (thenewstack.io) (9to5mac.com) Image generators turn written instructions into pictures; the new pitch is that this one spends extra compute on the request first, more like outlining before sketching. OpenAI has been pushing that “thinking” approach across other models, including o3 and o4-mini. (openai.com) (thenewstack.io) That changes the kinds of jobs the tool can do. OpenAI’s own materials and launch coverage emphasize layouts with dense text, edited uploads, transparent backgrounds, and production-style assets such as slides, maps, and magazine pages. (help.openai.com) (venturebeat.com) OpenAI says ChatGPT Images can take up to two minutes to generate a result, a tradeoff that signals the company is willing to spend more time per image to improve instruction-following. The editor also now supports any aspect ratio on web, iOS, and Android. (help.openai.com) The launch also lands with immediate safety questions because better text rendering and more consistent characters make it easier to produce polished fakes, parody ads, and impersonation-style images at scale. CNET reported that online stress tests started almost immediately after release. (cnet.com) (openai.com) OpenAI says it monitors and enforces its rules with automated systems and human review, and its policy pages include specific guardrails for image and video generation. The company’s broader safety page says it uses real-world feedback to update protections after launch, not just before it. (openai.com 1) (openai.com 2) (openai.com 3) The near-term test is less about whether the model can draw a better poster than last year’s version and more about whether OpenAI can widen access without losing control of what people make with it. (help.openai.com) (cnet.com)
