Dall-E 3 not spelling words correctly.

[verbal steppe]

The intent is clearly understood around "problem statements" yet the spelling in the image is incorrect.

Is this a known problem being worked on? are there any workaround to give explicit instructions for spelling?
I have replicated many times and wrong spelling is an issue.

Bug Report:

Steps to reproduce:

Prompt: Illustration of a lively brainstorming session. Participants are divided into small groups, each huddled around a table filled with sticky notes, markers, and whiteboards. Animated discussions take place as members jot down and highlight pressing problem statements.

Expected result:
Problem Statements

Actual result:
Problem Statetons

Additional information

Browser: Chrome
OS: Mac

[wintry zodiac]

It's the nature of the beast. Dall-E 2 basically couldn't do text. DALL-E 3 often can, but it varies a lot. A model trained to produce art is not as well trained to produce text. ChatGPT can explain more precisely why that is, but it will likely be awhile before DALL-E can reliably produce correct text in most images.

[verbal steppe]

Ok, I'd love for them to prioritise this, it's super helpful for me in my creative workflow.

I didn't ask Dall-e3 to put the text, I want it to produce accurate text or non at all, is there a way to command it not to produce text if that's not its strength given lack of training?

This way I can just produce a great illustration, and in my creative documents I'll produce titles, labels separately to the illustration.

As it stands the tool is in no mans land for my use case