OpenAI flips GPT-IMAGE-2 live, adds image/video generation to its GPT lineup

Image generation is now a normal OpenAI model line item — not a side project, not a rumor, and not a weird leaked label from a demo. The real news is simpler than the social chatter made it sound. OpenAI officially released GPT Image 2 on April 21, 2026, put it in the API docs, and surfaced it in developer tooling the same day. ### So what actually launched? GPT Image 2 is OpenAI’s current flagship image model for generation and editing. The official model page describes it as the state-of-the-art image generation model, with text and image input and image output, and the changelog pins the release date to April 21. ### Why does the rumor framing miss the point? Because this was not a stealth “went live today” moment on May 8 or May 9. (openai.com) The public breadcrumbs were already there — product post, model page, docs, changelog, and developer homepage placement. What people likely noticed in recent demos was a product rollout becoming more visible, not a secret model suddenly appearing out of nowhere. (developers.openai.com) ### What is GPT Image 2 supposed to be good at? OpenAI is pitching three practical upgrades — stronger instruction following, better text rendering inside images, and better editing. That sounds like marketing fluff, but it matters because older image models routinely fell apart on posters, diagrams, UI mocks, and anything with lots of labels. OpenAI also says the model supports flexible image sizes and high-fidelity image inputs, which makes it more useful for actual design workflows instead of just one-shot art prompts. (openai.com) ### Is this replacing the old image stack? Basically, yes. OpenAI’s docs now present “GPT Image” models as the image generation path, with GPT Image 2 as the latest option and GPT Image 1 listed as the previous model. That is a meaningful product shift — image generation is being folded into the GPT family naming scheme rather than living mainly under DALL·E branding. (openai.com) ### What about video — is that GPT Image 2 too? No. Video is still its own lane under Sora 2 and the Videos API. OpenAI’s current video docs point developers to Sora 2 and Sora 2 Pro for generated clips, while GPT Image 2 handles still images. So the “adds image/video generation to its GPT lineup” framing is only half right — OpenAI is broadening its media stack, but video is not being folded into a single GPT-IMAGE-2 model. (developers.openai.com) ### Is Sora fully live then? Sort of, but with a catch. The Sora web and app experiences were discontinued on April 26, 2026, while the Sora API remains available until September 24, 2026. So OpenAI still offers video generation to developers, but the consumer-facing Sora product has already been wound down. ### Why does this matter for builders? (developers.openai.com) Because OpenAI is making media generation feel like standard infrastructure. The API now has a current image model, a current video model, pricing pages, guides, and batch support. That turns creative generation from “cool demo” into something product teams can wire into workflows for mockups, marketing assets, edits, and short-form video experiments. (help.openai.com) ### What’s the bottom line? The story is not that a mysterious GPT-IMAGE-2 label suddenly appeared this week. The story is that OpenAI already shipped GPT Image 2, made it an official part of the platform, and now has a clearer split: GPT models for core multimodal work, GPT Image 2 for images, and Sora 2 for video — at least for as long as the Sora API stays up. (openai.com) (developers.openai.com)