Hey all,
I’m trying to make some potential logos, but text is always included (and as we know Dalle text tends to be gibberish). As much as I say, “The picture has no text” or “there are no words in the logo” Dalle always seems to want them in there anyway. Does anyone have prompt strategies for excluding text?
Negative prompts are tricky, so I would recommend not mentioning the word “word”, and instead switch the word “logo” to “symbol”, “emblem”, or “icon”, as those are less closely related to “text”
What’s your prompt look like so far?
“A simple, icon for a company that builds AI models for companies to give their data a better view”
(slightly paraphrased to keep some things private)
I’ve switched to Icon because that provides slightly better results (1/4 doesn’t have text)
Might add “digital art”… “marketing materials” … “simple icon…”
Good luck!
Well that didn’t work haha.,… the graphic icon on the far right isn’t bad, tho…
I’ll see if I can come up with a better solution and get back to you…
You’re right it’s not bad! Thanks for taking a crack at this!
I think it’s taking the words “logo” and “icon” to heart, so maybe try without…
This person went through something similar, it seems…
ETA: Although he did use “logo” in the later attempts… I wonder if putting what you want (which is hard because you want the creative ideas) in front of the prompt and logo / icon at the very end…I’m pretty sure stuff at the start of the prompt gets valued more, but I may be wrong…
ETA2: Yeah, looks a little better… I would experiment with word placement…
Did anyone figure this out? I dont want text im my images either.
No dice unfortunately. I ended up just using the editor and making a million versions until I got something I could use.
I need help on this too since im doing this programmatically I cannot view the images each time
Just started using this, guess I’m not alone with this issue. It seems like if you ask for something that is a found object like cat or car it’s no problem, but when it’s something more abstract like “economic recession” it always includes some random letters
I used someones suggestion to avoid negative prompting and gave a bit more instruction on what I actually wanted - even though it is abstract. I’m also trying to do it programmatically so having some good prompt ideas would be helpful.
“create a square business logo containing simple colours and shapes inspired by artificial intelligence”
I didn’t get any text
Also I tried to make it a bit less abstract by putting something concrete in the middle of the idea and the image.
“a 2D pencil drawing that looks like a scene from an animated movie about diversity and inclusion”
Weird images but not much text
Yeah I would have never thought of that of course I’m not like a computer or anything but I am learning this stuff and have been using it for my own projects and that seems to make sense what you said.
Just adding to the consensus here. So far I’ve found OpenAI incredible, but Dall-E outputting words/text/characters just seems to make it completely unusable IMO? If there’s a reason for it (such as copyright etc), then I guess that’s fair enough. But if not - I’d say it would be good to change this. Is it a bug? Can’t see why it shouldn’t accept exclusions.
The text completions don’t have any trouble with negative/exclusions in prompts (that I’d noticed - I could be wrong) Curious why the image one does?
I have the same issue all the time. I wish it didn’t ever try to include text–it shouldn’t because it can’t ever do that reliably–and no one even wants it to. We can create word art and text in so many other ways.
For logos, there are a lot of other options out there to check out. Logo maker AIs, go check them out instead.
Does anyone have any insight as to why DALL-E includes meaningless jumbled text like this?
It seems that Bing gets the most advanced version of Dall-E. I’ve never seen inclusion of text in something like logo when described.
“Simple original geometric shape two color company logo on white background with vector graphics that subtly reflects the company’s development of AI products. electric green and neon purple as primary colors.”
That it can’t write text is due to its training on the semantics of imagery, the human tagging of “pick all images with pictures of a rat”, and the neural vision knowledge that puts those elements together. A prompt, which I know would fail due to this, actually gives cool ideas, synthesizing similar concepts:
"A band flyer advertisement, with a few guitar and music graphics. Text to include: “Live Saturday at the Rustic Pub”. “The Rabid Badgers” band. “9-midnight”. “$10 cover, all ages”.
My avatar image came out of it.
