That tip about adding tiny details to the prompt was dead wrong for mashups

My buddy Rick who does digital art kept telling me to stack every prompt with super specific stuff like lighting and texture words. He said it makes AI pics look more real. So I tried it on a mashup of a lighthouse made out of waffles and syrup. I typed like 40 different detail words and the thing came out looking like a blurry mess with weird faces in the clouds. Then I stripped it down to just "lighthouse, waffle texture, syrup waves" and got the funniest result I've made all month. Turns out for mashups, the silly concept carries the image and extra detail words just confuse the model. Has anyone else found that shorter prompts work better when you're mixing weird things together? I'm curious if it's just me or if the AI gets too focused on making everything perfect.

1 comments

Log in to join the discussion

Log In

1 Comment

brian_rivera5924d ago

I saw a Reddit thread where a guy tested 10 prompts on the same weird concept and the short one won every time. His guess was that too many words make the model try to satisfy all of them at once, so the main idea gets watered down. Your waffle lighthouse is basically proof of that, the dumb idea is the whole point and the details just get in the way.

5