DALL-E 2 and other image generators are well known for this specific effect on text. It has even been shown that the choices of "words" that DALL-E uses are subject-specific, with the same kinds of mis-spellings appearing again and again.
This is a specific type of failure of coherence of the model. It's not really that different from the generator mangling hands, positioning limbs incorrectly etc. There isn't really anything special about text to (this kind of) AI in this regard.
Some versions of Midjourney (version 3, there have been a few versions since, but version 3 is still available and the default for a few weeks yet) have a better, but still not reliable, response to generating text. You may have some luck there.
Other than that, it's a waiting game. There are state-of-the-art models that can reliably create text, even text with artistic effects, more reliably. This seems to relate to the power of the model. Google's research project Parti published an article to demonstrate this effect of adding more parameters to the model. Their best model generates very good-looking text. Google's other model Imagen does too. The main problem for you being that these models are not available to the public. At least not yet.
My advice: Generate your artsy background in the image generator, where the specific locations of components matter less. Then use an image editor to add titles in a style of your choosing. Fighting with the generator to get working text will take far more effort than that.
