Does anyone experience issues with Dall-E3 generating typos in text within images?

I have been using Dall-E3 to help my girlfriend with a project and she needs some text in the image because it is a logo with some small text.

We have encountered numerous times that out of 100 logos, only about 10 of them have the correct spelling as requested word for word. Then when you ask Dall-E3 to keep the image intact but to fix the typos, it changes the images into completely different images with spelling errors.

Has anyone been able to circumvent this issue with other prompting?


But for the effort of having made it do text better - today’s DALL-E 3:


Have you found any way to get around this or for it to be not so severe that you can change it later on with paint or something similar?


Yeah, I get that with every model, the text-to-image and text-to-video models do it, all, except for Adobe’s Firefly. Probably because Adobe fine-tuned their model with all those fonts, they “own”? YMMV.

Yes, same for me. It’s very annoying!

Yes, and in frustration, I’ve managed to get it to come out with some great swearing that it would otherwise reject!

Although that said, I’ve noticed it’s improving.

It’s so bad lol. I always specifically request in the prompt to use no words or letters, to be visuals only. And it still puts out words in the output.

I’ve had problems with this… IT comes down to other words hinting that there’s text/numbers in the image. Hard to squash completely, but I’m getting better…

1 Like

same here as well… even if I instruct it to only use words that appear literally in the input prompt , still happens… would be great if one way or the other this could be solved

1 Like


I have same problem, any text in Vietnamese got broken and making image unusable.

Anyone has idea with Unicode text?

From my personal experience I feel like the last update kinda made GPT less good at typing (personal feeling). With Dalle-3 first version i was having a good amount of images with correct text: it’s still possible now, but will require more attempts. Finger crossed we gonna have updates about text.

From what i found: short text tend to works better than longer text.
Putting the text in front of the prompt might gather better results.

It would be great to see the underlying image generation without spelling mistakes in an upcoming update from OpenAI. I’m creating a custom GPT with this specific use case.

1 Like

I encountered this problem recently where it could not spell the word “PSYCHOLOGY” correctly. After five attempts and inclusion in the prompt instructions to spell the word correctly, I had one success though I didn’t like the image as well.

I complained to it and was informed that it’s a technical challenge that people are working on. The Dalle does not create with language but aesthetics. It ‘s utilizing , processing and creating images only. I was not happy about it:(.

1 Like

I have noticed for several weeks that Copilot and, as a result, DALLE have great problems with correctly entering texts into graphics - there are numerous spelling errors, duplicate letters and inserting additional characters without the need.

1 Like