How do I prompt Dall-E to include specific sentences in image creation and not misspell the words and sentences provided?

thank you.

1 Like

Heya. Welcome to the developer forum.

Can you share the prompt you used?

A lot of words still isn’t easy to do, though it’s improving.

Share your prompt, though, and we’ll see if we can help.

1 Like

Oh, sure, still new. Here is my prompt… Create a fun square shaped visual mural based on a travel theme of Spain as the background.
Include within the mural the following words and sentences, precisely and exactly as typed between the quotes, separating by commas; “HI IBIZA”, “MOONLIGHT JAZZ”, “BOSPHORUS”, “POETRY SLAM”, “SANGRIA FAMILIA”, “DO YOU KNOW HOW TO MAKE IT BIGGER?”, “IS THAT MY TOWEL?”, “SIESTA BEFORE FIESTA”, “KENOUGH”, “ISTANBUL”, “IBIZA”, “DAVID GUETTA”, “UNO”, “SEXY STRONG INDEPENDENT LATINA WOMEN”, “BARMUT”, “WHO ARE THEY?”, “PLACEHOLDER”, “BARCELONA”, “JUST ONE DRINK”, “THE EAGLE HAS LANDED SEVERAL TIMES”, “THEY MUST BE ITALIAN”

Yeah, that’s a bit out of it’s capabilities at the moment.

Try starting with one or two phrases… THEN you can “edit” the image in ChatGPT to begin adding other phrases.

Let us know how it works!

1 Like

Ah, so that uses the regenerate function to do an “edit?” Is that essentially "edit and add the following sentence to the scene? or something like that?

1 Like

oh, nevermind, I see the edit function.

1 Like

But I shall challenge the might of Dalle 3 through API and have as much as possible covered even if it takes 100 USD worth of credits. :sweat_smile::wave:

Shall be. Fun little activity Sir wanna ?

You might try selecting the entire image and asking it to correct it’s spelling errors. I’ve seen that work on images before.

Meanwhile, Paul is right. All of the image generators have trouble with text. Based on what I’ve seen, what you got is pretty awesome, despite those spelling errors. That text is all very distinct and reproducible.

If this is something important, I suggest bringing the image into an Adobe Product and working with actual fonts—Adobe has some font-identifying abilities that can help you figure out which font Dalle had in mind.

Thanks thinktank. I’ll try those suggestions. Editing in place and trying to get it to insert “text, sentence, or phrase” just doesn’t work. But I like the idea of getting it to correct spelling. I’ll try that. I might also try exporting image to another product if I can’t figure out prompts for this. Thanks!

1 Like

To ensure that DALL-E includes specific sentences in an image and spells them correctly, follow these guidelines:

1. Be Explicit in Your Prompt

Clearly state that the text must be included exactly as provided, and specify that accuracy in spelling and punctuation is crucial.

2. Use Simple and Direct Language

Avoid overly complex sentences. Clearly list the sentences or words you want to be included.

3. Break Down the Request

If necessary, break down your instructions into smaller parts to ensure that DALL-E understands each element. For instance, you can specify the background and then list the text separately.

Example Prompt

“Create a fun, square-shaped visual mural with a travel theme of Spain as the background. Include the following words and sentences exactly as typed, with correct spelling and punctuation: ‘HI IBIZA’, ‘MOONLIGHT JAZZ’, ‘BOSPHORUS’, ‘POETRY SLAM’, ‘SANGRIA FAMILIA’, ‘DO YOU KNOW HOW TO MAKE IT BIGGER?’, ‘IS THAT MY TOWEL?’, ‘SIESTA BEFORE FIESTA’, ‘KENOUGH’, ‘ISTANBUL’, ‘IBIZA’, ‘DAVID GUETTA’, ‘UNO’, ‘SEXY STRONG INDEPENDENT LATINA WOMEN’, ‘BARMUT’, ‘WHO ARE THEY?’, ‘PLACEHOLDER’, ‘BARCELONA’, ‘JUST ONE DRINK’, ‘THE EAGLE HAS LANDED SEVERAL TIMES’, ‘THEY MUST BE ITALIAN’. Ensure the words and sentences are spelled and punctuated exactly as provided.”

Additional Tips

  • Preview and Edit: If the initial output has errors, rephrase your prompt for clarity and run it again.
  • Specify Font or Style: If the design of the text matters, include instructions about the font, size, or placement.
  • Check Examples: Look for examples or templates that DALL-E may have used successfully to incorporate text accurately.

@aliboukaroui , thanks, my original prompt was pretty close to your recommended prompt, but I also tried your recommended prompt, unfortunately, same result. Decided to just update prompt to create well defined quote bubbles for text, then export to photoshop to finish project. Thanks all.

3 Likes

I’ve experienced asking it to “edit” a misspelled word and then the response was an image without the word altogether and the chat told me it made the suggested edit even though the word was removed altogether.

What I’ve done for a similar project is export to Adobe Illustrator, then use illustrator’s Retype to identify a font that approximates what Dalle had in mind.

The only problem with the image you shared is the hallucinated text, not the font choices.

Please share your final version! I’d like to see how you bring it all together.

1 Like

Happy to share final. Full disclosure, I am not an artist or adobe expert, “fixing” the misspelling of SPAIN in the center middle was outside my skillset and I couldn’t get spelling correction prompts to work (results were worse). I did generate a different image specifying I wanted quote bubbles for the text, then was able to get that into photoshop, found a font that reasonably matched, and got something workable. As a non creative who loves being creative, being able to put something together like this in a few hours is simply amazing.

2 Likes

That’s the spirit! We’ve still got a few months (years) or human+ai hybrid work, but that time will pass eventually.

Glad you got it sorted.

1 Like