OpenAI launches ChatGPT Images 2.0: Major evolution in text rendering, supports multi-image generation, but "Chinese generation" remains a hurdle

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1394 words

AI image generation evolves again! OpenAI officially launched its brand-new image generation model, "ChatGPT Images 2.0," this Tuesday. The new model significantly enhances "reasoning capabilities" and "text rendering" technology. Users can now generate multiple images at once, customize extreme aspect ratios, and even accurately generate English words within images. However, foreign media tests found that the model still produces unrecognizable "AI gibberish" when handling non-English languages like Chinese. (Previous coverage: A lifespan of only 3 months! OpenAI unexpectedly shuts down its research writing platform, Prism, shifting strategy to "no more side projects.") (Background supplement: ChatGPT key figure Srinivas Narayanan abruptly resigns from OpenAI; three executives have left in one week.) ChatGPT Images 2.0, with a more powerful computing version provided for paid subscribers. The battlefield for image generation is heating up again. OpenAI announced on Tuesday (the 21st) the launch of a brand-new image generation AI model for global ChatGPT and Codex users. This major update not only brings more detailed visual performance but also attempts to solve the most troublesome issue in AI image generation: "Text rendering." Combining reasoning capabilities, a single prompt can produce multiple images. Compared to previous models, the biggest breakthrough of Images 2.0 is its integration of ChatGPT's powerful "reasoning" capabilities. This means that before generating an image, the AI performs more thought steps and can even connect to the internet to search for the latest information (the model's base knowledge cutoff date is December 2025). Highlights of the new model's upgrades include: - Continuous multi-image generation: Users only need to input a prompt once to have the model produce a series of images, such as the visual content for an entire study manual. - Highly customizable dimensions: Breaking traditional aspect ratio limits, the new model supports aspect ratios from 3:1 (extremely wide) to 1:3 (extremely tall), which users can specify directly in the prompt. - More detailed infographics: When foreign media tested the model by asking it to generate an infographic for "San Francisco's tomorrow weather forecast and recommended activities," the AI successfully integrated weather details and local landmarks (such as the Ferry Building, Castro Theatre, and Transamerica Pyramid) accurately into a single image. English spelling passes perfectly, but "Chinese posters" turn into gibberish. In the past few years, when mainstream models attempted to generate text within images, they often produced distorted characters or misspelled words. According to tests, Images 2.0 has made stunning progress in English text rendering, with English words in the images becoming much clearer and more accurate. However, when challenged with non-English languages, Images 2.0 still struggles. Foreign media testers asked ChatGPT to imitate Chinese fans and create a "Chinese support poster" for Hollywood actor Timothée Chalamet. Although the resulting poster was visually striking (including elements like traditional clothing, cat ears, bubble tea, and pandas) and filled with over 20 pieces of text, the characters were unreadable. When testers asked ChatGPT what the text meant, the AI displayed strong "self-critical" capabilities, honestly replying: "Most of this is fake, or semi-gibberish AI text disguised as a Chinese meme poster, so it cannot be translated fluently. There are also places that are clearly distorted or mixed with characters that look like Japanese... These are mostly meaningless symbols fabricated to mimic the feel of East Asian fan-edited text, rather than accurate sentences." In summary, ChatGPT Images 2.0 has already demonstrated powerful capabilities in functional diversity and English processing, undoubtedly bringing substantial improvements to productivity tools. As for the "accurate multi-language generation" that global users are eagerly anticipating, it may still require waiting for OpenAI to strengthen it with more extensive global data in future versions.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching6 items

2026-04-23

OpenAI launches ChatGPT for Clinicians, assisting human physicians with 99.6% accuracy (free access in the US)

Similarity 170%關鍵字 openai/chatgpt同分類 zh

2026-04-20

Srinivas Narayanan, a key figure behind ChatGPT, has abruptly resigned from OpenAI, marking the third high-level departure in a week.

Similarity 170%關鍵字 openai/chatgpt同分類 zh

2026-04-22

OpenAI Launches Workspace Agents Feature in ChatGPT

Similarity 130%關鍵字 openai/chatgpt

2026-04-22