Mr. Minh Quan - owner of an online store in Hanoi, said that in the past, when creating images with ChatGPT, the text often had incorrect accents, deviated fonts or was difficult to read.
Every time I create photos to post for sale, from banners to product photos, the text part almost has to be completely corrected because of incorrect accents, off-key fonts or difficult to read. Some days I spend an hour just adjusting the text, so it takes even more time than self-designing. So I almost don't use ChatGPT for this job anymore" - Mr. Quan said.
Because of these limitations, Mr. Quan as well as many other users gradually switched to other tools such as Gemini of Google when they needed to create photos with words. Compared to ChatGPT before, the text display capability of this tool is considered more stable.
However, this experience changed after ChatGPT's Images 2.0 update. When trying again to make banners, Mr. Quan said the results were "no different from before". Vietnamese characters are displayed clearly, correctly marked and can be used immediately without too much editing.
Ms. Thu Hang - a graphic designer in Hanoi said that initially she only tried again to know, but the results surprised her. "I do infographics so I am very afraid of typographic errors. Before, I almost didn't use ChatGPT because I had to edit too much. This time creating it for testing, the letters are mostly correct, the layout is also maintained, just adjust lightly and it can be used," Ms. Hang said.
Meanwhile, Mr. Duc Anh - content creator on social networks, also expressed surprise when experiencing the new Images 2.0 update. "I tried to make a series of quoted photos in Vietnamese and the returned results showed quite well, I can use them now without much editing," Mr. Duc Anh shared.
According to OpenAI, Images 2.0 is improved to better handle small details such as text, icons and complex layouts. The model is also integrated with reasoning capabilities, allowing self-checking and result optimization before display, thereby reducing errors in difficult components such as writing.
AI accurately displaying text is a major challenge due to the specific nature of diffusion technology, which often prioritizes reproducing overall images rather than small details. This makes writing, especially accented languages such as Vietnamese, easily distorted.
Practical testing shows that the new version has significantly improved this limitation. Long text still retains clarity and spelling correction in many cases.
However, the tool is still not completely stable. Some complex layouts or information-rich charts may still have errors.
From being considered "not worth using" with Vietnamese, ChatGPT is gradually regaining user trust. The ability to display text accurately is considered an important factor determining whether AI can go into practical applications or not.
