News for hackers. Stuff that matters.
Re:DALL-E 3: Improving image generation with better captions... by astrange on Thursday October 19, 2023 @10:50PM (#37951848)
DALLE3 is explicitly designed around not having to do this; the point is that it'll write the detailed prompt for you in four different ways so you get more variation.

"Wide angle" and "fisheye" do work, but "lens" is dangerous; anything that can be read as an object will tend to cause that object to appear instead of being used metaphorically.

> - Again, be more descriptive and specific with your prompts. ie; 'a character in the style of 'Naruto'. Need to specify the show or artists instead of using the broad term "anime".

Explicitly against the rules. It'll block you if you try to use any real person's name or anything it thinks is copyrighted. You can paraphrase though, or argue with it which weirdly sometimes works.

> One more time, you just need to be more specific. do: 'a computer keyboard, but instead of regular keys, replace each one with a metal house key.'

Tried it, doesn't work. If it gets to dalle it's not smart enough to reliably do "instead of" (or generally anything that's a "deletion"). It'll just put in all three concepts.

Synonyms can help if there is one in English. Emoji do interesting things but can't tell how well they work.

There are no comments on this comment yet.