Took me 3 months to realize my AI art prompts were too specific

I kept getting these stiff, lifeless images from Midjourney no matter what I tried. Every prompt was like 'a woman standing in a forest at golden hour with perfect lighting and a flowing dress and a gentle smile'. They looked polished but totally dead. Then I saw this guy in the Discord post something super vague like 'lonely street corner, rainy night' and his results had way more soul than mine. It hit me that I was over-directing every tiny detail and leaving no room for the AI to actually interpret anything. So I tried a prompt that was just 'old bookstore cat' with no adjectives and the weird imperfections made it feel like actual art. Has anyone else had this moment where less detail made way better results?

2 comments

Log in to join the discussion

Log In

2 Comments

derek9393mo ago

Yeah the 'gentle smile' part got me because that's exactly what I was doing for months. Every image looked like a stock photo of a robot at a Renaissance fair. I think the problem is we get so caught up in trying to control everything that the AI just goes into autopilot and gives us the blandest possible version. Honestly my best results came from typing 'angry raccoon chef' with zero punctuation and letting it cook.

5

spencer_ross3mo ago

Last week I typed "sad potato accountant" and got the most beautiful image I've ever generated.

3