ChatGPT images 2 boosts realism
- OpenAI said on April 21 it rolled out ChatGPT Images 2.0 in ChatGPT, plus the underlying gpt-image-2 model in the API and Codex. - OpenAI says the model can generate up to eight images per prompt, render multilingual text, and handle posters, slides, maps, and UI mockups. - The release extends OpenAI’s push into design workflows already tied to Canva inside ChatGPT. (openai.com)
OpenAI rolled out ChatGPT Images 2.0 on April 21, adding a new image model to ChatGPT and releasing gpt-image-2 in the API and Codex. (openai.com) (community.openai.com) Image models turn text prompts into pictures, but earlier versions often mangled letters, broke layouts, or changed characters from frame to frame. OpenAI says Images 2.0 was built to follow longer instructions, place text more accurately, and keep visual elements consistent across multiple outputs. (openai.com) (developers.openai.com) OpenAI’s product page says the model can generate up to eight images from one prompt and supports multilingual text rendering. The company’s examples show magazine covers, maps, slides, manga panels, and infographic-style layouts. (openai.com) For developers, the same model is now available through OpenAI’s image generation tools and API docs under the name gpt-image-2. OpenAI’s developer announcement said the model is aimed at “complex visual tasks” with stronger editing, better layouts, and more reliable instruction following. (developers.openai.com) (community.openai.com) The release targets a problem that has limited AI images in real work: a poster or user-interface mockup fails if the headline is unreadable or the buttons drift. OpenAI’s own prompting guide frames gpt-image-2 as a production tool for design systems, marketing assets, and edited image workflows, not just one-off art. (developers.openai.com) (openai.com) Arenas that rank image models by head-to-head user preference also gave the launch extra weight this week. Arena’s text-to-image leaderboard shows gpt-image-2 at the top of its overall rankings, though leaderboard positions can shift as more votes come in. (arena.ai) The Canva angle is less future tense than some early demos suggested. Canva already has an app inside ChatGPT that lets users generate, preview, and edit Canva designs in a conversation, and Canva’s help pages describe creating designs from prompts after connecting an account. (chatgpt.com) (canva.com) That puts Images 2.0 inside a broader race over whether chatbots can become design tools with exportable, editable output. OpenAI’s bet is that better text, tighter layouts, and multi-image consistency make image generation useful in places where older models still looked like drafts. (openai.com) (developers.openai.com)
