OpenAI Upgrades ChatGPT Image Generation
▼ Summary
– OpenAI released ChatGPT Images 2.0, a new model that can generate multiple images and text, including in non-English languages, from a single prompt.
– The model leverages ChatGPT’s reasoning to search the internet for recent information and produce more thorough, granular outputs, like a detailed weather infographic.
– It features a more recent knowledge cutoff date of December 2025 and offers users customizable image aspect ratios, from wide to tall formats.
– The release has improved text rendering in images, addressing previous issues with malformed characters and inaccurate labeling seen in older models.
– Major AI model launches can boost platform usage, especially if they spark social media trends, as seen with past releases from Google and OpenAI.
A significant upgrade to AI image generation arrived this week as OpenAI introduced ChatGPT Images 2.0. This new model enables users to create multiple images from a single prompt, such as a complete booklet, and can integrate text in various languages including Chinese and Hindi. The tool is now accessible worldwide for ChatGPT and Codex users, with a more advanced tier offered to paying subscribers. Major model releases often spark renewed public engagement, particularly when they enable viral social media trends. Last year, Google’s Nano Banana model captured widespread attention as people shared hyperrealistic digital figurines of themselves. Similarly, earlier this year, the original ChatGPT Images feature fueled a wave of AI-generated caricatures across platforms.
The latest iteration distinguishes itself by leveraging ChatGPT’s underlying reasoning capabilities. This allows Images 2.0 to pull recent information from the web and produce several coordinated images simultaneously. By employing additional internal steps, the model delivers more comprehensive and detailed outputs from one user instruction. A key technical improvement is its updated knowledge cutoff date of December 2025, ensuring responses reflect more current information.
These advancements result in outputs with greater granularity and accuracy. For instance, requesting an infographic for San Francisco’s next-day weather and suggested activities yielded a detailed image. It correctly depicted a rainy forecast alongside recognizable sketches of local landmarks like the Ferry Building, the Castro Theater, the Painted Ladies, and the Transamerica Pyramid.
The new model also offers enhanced customization for aspect ratios. Users can now specify dimensions within their prompts to generate images in formats ranging from a wide 3:1 layout to a tall 1:3 frame, providing greater flexibility for different visual applications.
In practical testing, the model’s text rendering capabilities show marked improvement, at least for English. Previous generations of leading image models, including ChatGPT’s own tools from just two years ago, frequently produced garbled text with malformed characters or extra letters. The cleaner and more complex textual elements in Images 2.0 outputs signal meaningful progress in this challenging area. This focus on legible text aligns with similar efforts by competitors like Google in recent versions of its Nano Banana model.
(Source: Wired)
