GPT Image 2 nails multilingual text
- OpenAI said on April 21 that ChatGPT Images 2.0 and the new gpt-image-2 model improve text rendering inside images, including multilingual layouts, and are rolling out in ChatGPT, the API, and Codex. - OpenAI’s prompting guide says gpt-image-2 delivers “reliable text rendering” with crisp lettering, consistent layout, and structured visuals such as infographics, diagrams, and multi-panel compositions for production workflows. - The release extends last year’s push beyond GPT-4o image generation by targeting practical design work that earlier models often mangled with broken typography. (openai.com)
Image generators have long been good at pictures and bad at words. OpenAI says its new ChatGPT Images 2.0 and gpt-image-2 model fix more of that problem, including multilingual text inside layouts. (openai.com) OpenAI announced ChatGPT Images 2.0 on April 21, 2026, calling it a state-of-the-art image generation model with improved text rendering, multilingual support, and stronger visual reasoning. (openai.com) The company said gpt-image-2 is available the same day in the API and in Codex, extending the release beyond ChatGPT to developers building image tools and automated workflows. (community.openai.com) (developers.openai.com) The technical change is simple to describe: older image models often treated text like decoration and produced misspellings, warped letters, or gibberish. OpenAI’s new model is being pitched for cases where the words in the image are part of the product. (openai.com 1) (openai.com 2) OpenAI’s own prompting guide highlights “reliable text rendering,” “consistent layout,” and “strong contrast” inside images, and points to infographics, diagrams, and multi-panel compositions as target use cases. (developers.openai.com) That matters for multilingual work because translation is not enough if the final poster, menu, comic panel, or user interface mockup still breaks when the text changes scripts. OpenAI’s release explicitly lists multilingual support as part of the upgrade. (openai.com) The release also shows how OpenAI is repositioning image generation away from novelty art and toward production design. Its API docs describe GPT Image models as tools to generate or edit images from prompts and image inputs, with gpt-image-2 presented as the latest option. (developers.openai.com 1) (developers.openai.com 2) This is the second straight year OpenAI has framed text-in-image accuracy as a core benchmark. In March 2025, the company said GPT-4o image generation could “reliably” incorporate text into images; the April 2026 update adds multilingual support and pushes the capability into a newer model family. (openai.com 1) (openai.com 2) OpenAI has not published a public benchmark in the release materials that scores Hindi, Japanese, Arabic, or other scripts against rival models. The evidence in the official materials is descriptive: better text rendering, multilingual support, and examples aimed at structured design work. (openai.com) (developers.openai.com) So the cleanest claim is also the narrowest one: OpenAI has launched a new image model that says it can place multilingual text more accurately where businesses actually need it — inside the image itself. (openai.com)
