OpenAI Unveils 'ChatGPT Image 2.0': "No More Broken Korean Text"

by Lee Eunseo

Published 22 Apr.2026 14:57(KST)

The Identity of "Duck Tape" Unveiled

OpenAI released a new version of its artificial intelligence (AI) image generation tool, "ChatGPT Image 2.0," on April 21 (local time).

Image created with ChatGPT Image 2.0, naturally rendering multilingual text. OpenAI.

This service, built on the "ImageGen 2.0" model, is the official launch of the tool previously known as "Duck Tape" (tentative name), which attracted attention after being highly praised by users of the AI evaluation platform Arena for almost perfectly resolving the challenge of rendering text in images.

Previous AI image generation models often produced results with meaningless arrangements of consonants and vowels or illegible, distorted text. Even when asked to correct the text, it often took a long time or resulted in even worse outputs. However, Duck Tape was evaluated as eliminating the sense of awkwardness unique to AI-generated images by accurately rendering Hangul within images. Users shared images created with Duck Tape on online communities, expressing positive reactions and calling it "innovative" for even making the so-called "alien language" disappear.

This new model precisely reflects users' detailed instructions, providing more versatile outputs than previous models. It can accurately arrange the positions and relationships of objects within images, and delivers improved results in challenging areas such as small text, icons, user interface (UI) elements, dense layouts, and style constraints. It supports aspect ratios up to 3:1 or 1:3, and can reproduce various styles, including photography, cartoons, and film.

Text rendering quality has also been improved, enhancing multilingual capabilities not only in Korean but also in Japanese, Chinese, Hindi, Bengali, and more. Notably, this is OpenAI's first image model based on "ChatGPT Image Reasoning." It supports information retrieval via web search, the generation of multiple images from a single prompt, and result inspection functions.

The tool is available in ChatGPT and Codex, while advanced output features based on ChatGPT Image Reasoning are offered to ChatGPT Plus, Pro, and Business users.