OpenAI on the 21st (local time) unveiled ChatGPT Image 2.0, a next-generation model that further strengthens image generation.
The service was developed based on ImageGen 2.0. It is the official release version of Duct Tape, which was evaluated as having significantly improved text rendering issues in user tests on the AI evaluation platform Arena.
ChatGPT Image 2.0 improves the text quality that had been cited as a previous weakness. Text rendering accuracy has increased across languages including Korean, Japanese, Chinese, Hindi, and Bengali, and it can now render small text.
Image resolution is supported up to 2K, and it can generate in a variety of aspect ratios from 3:1 to 1:3. It can also create up to 10 images at once.
It can also reproduce specific styles, including icons as well as comics and films. OpenAI said it can deliver more useful results than before by precisely reflecting users' detailed instructions.
Meanwhile, Thinking and Pro models, which enhance reasoning during image generation, were released alongside it. Using them can yield more accurate results and maintain consistency of people or characters across multiple images such as comics.
OpenAI said it applied digital watermarks such as SynthID to help identify whether content was generated by AI in order to prevent misuse.
ChatGPT Image 2.0 is available to all accounts, including free users. However, advanced reasoning-based output features are offered only to paid users such as Plus, Pro, and Business.