📰 AI News: ChatGPT Can Now Think Before It Draws, and That Changes What AI Images Are Good For
📝 TL;DR OpenAI just launched ChatGPT Images 2.0, and it looks like a real step change, not just a prettier image model.The big upgrade is that it can reason through more complex prompts, render text far more accurately, and generate consistent image sets that feel much closer to production-ready creative work. 🧠 Overview OpenAI’s new Images 2.0 model is now rolling out across ChatGPT, Codex, and the API as gpt-image-2. The company says this version improves instruction following, multilingual text rendering, layout control, and editing for complex visual tasks like infographics, comic pages, posters, slides, and UI mockups. The bigger story is that AI image generation is shifting from “make me something cool” to “make me something usable.” 📜 The Announcement OpenAI says ChatGPT Images 2.0 is available on all ChatGPT plans, while “images with thinking” is limited to paid plans through the Thinking and Pro model options. In the API, developers can use gpt-image-2 starting now, and OpenAI describes it as its most capable image generation model yet. Reporting also says the new model can search the web, generate multiple consistent images from a single prompt, and double-check outputs before finalizing them. ⚙️ How It Works • Thinking before generating - The model can spend extra time planning and refining outputs before it creates them. • Better text rendering - The new model is much stronger at small text, labels, iconography, UI elements, and other details that used to break image generation. • Multilingual support - OpenAI highlights stronger rendering for non-Latin scripts including Japanese, Korean, Hindi, and Bengali. • Multi-image consistency - It can generate multiple images from one prompt while keeping characters, style, and objects more consistent across scenes. • Flexible outputs - OpenAI showcases everything from manga pages and academic posters to panoramic scenes and print-ready layouts, with support up to 2K resolution.