Korean Expat Community for National Disasters

• 목록

For a while, I mainly used Gemini for AI image generation. It was perfectly adequate for general images and worked reliably for most purposes. However, I ran into problems every time I tried to generate images that included text. Letters would come out garbled, blended together, or just completely wrong. This wasn't a minor issue — it made the images unusable for anything that required actual legible text.

Editing the text separately in the generated image wasn't a good solution either. I tried fixing it in other tools, but it added extra steps and the results rarely looked clean. Eventually, I just accepted it as a limitation of AI image generation — text in images is hard, and most tools don't handle it well.

Then I saw that ChatGPT had upgraded its image generation feature, so I decided to test it on exactly the kind of task that had been giving me trouble. I had a task where I needed to change specific text in an image, so I ran it through ChatGPT.

The difference was immediately apparent. The text came out clean and legible — no garbled characters, no blending. The words were exactly what I had specified, positioned correctly, and readable at a glance. It wasn't even a complex request, but a task that kept failing in Gemini worked cleanly in ChatGPT.

I also tested modifying text in existing images. The changes were applied naturally without distorting the surrounding image or blurring the text. I could use the results immediately without any additional cleanup in another tool.

Of course, Gemini still has its strengths. It handles images without text well and produces good results in certain styles. But for image tasks where text needs to be accurate — posters, thumbnails, labeled charts — ChatGPT's upgraded image generation is clearly better. It was the first AI image tool where I didn't have to budget extra time fixing broken text after the fact.