ChatGPT Images 2.0 upgrades

OpenAI rolled out ChatGPT Images 2.0 on April 21, adding sharper image generation, better text inside images, and support for multiple image inputs. (openai.com) OpenAI said the new model improves multilingual text rendering and visual reasoning, and the company added an “images with thinking” mode on paid plans that lets ChatGPT spend more time planning an output before it draws it. (openai.com) (help.openai.com) The company’s system card says the model can pull in live web search data, generate several images from one prompt, and turn a basic request into a more detailed image brief using OpenAI’s reasoning tools. (deploymentsafety.openai.com) OpenAI is also pushing the upgrade into developer products. The company said gpt-image-2 is available in the API and Codex, where it is aimed at production work such as layouts, editing, and images that need legible text. (community.openai.com) (developers.openai.com) That changes where image models fit inside software. OpenAI’s API docs say image generation can now take prompt text plus image inputs inside a conversation, which gives app makers a way to build design, editing, and mockup tools without switching systems. (developers.openai.com) Reviews published on April 22 said the jump was visible in side-by-side use. MacRumors wrote that the update improved realism and text handling, while PCMag said web access and multi-image support made prompts more useful for composite scenes and richer context. (macrumors.com) (pcmag.com) The same improvements widen the abuse case for fake visuals that look like screenshots, ads, or news clips. Business Insider reported that stronger text rendering makes it easier to fabricate convincing image-based hoaxes, including fake article layouts. (businessinsider.com) OpenAI says generated images include C2PA metadata, an industry standard for provenance that can help verify where media came from. The company’s help center says people can check that metadata with verification tools, although metadata can be stripped when files are reposted or edited. (help.openai.com) OpenAI’s broader usage policies still ban harmful image misuse, and the company has tied image provenance to its safety work since 2024. The question after this release is less whether the pictures look better than before, and more how often viewers will know where they came from. (openai.com 1) (openai.com 2)