Everyone knows how to prompt an image. Not everyone gets the image they actually had in mind. That was my problem: the AI images I prompted looked good, but they weren’t right.
So I went to a workshop on AI images to find out why. I’ve split this article into what I took away myself and the specific methods I learned at the workshop, so it’s clear which is which.
What I took away
It’s not just about prompting it right. It’s the thought process and the structured thinking you bring while you prompt.
It’s like photography. You need good timing and good lighting to get a good shot. With AI images, you need a good prompt and a good structure to make the image look right.
The biggest takeaway for me: I can now look at an image, identify what’s wrong with it, and know how to fix it.
Methods I learned at the workshop
These come from the workshop rather than from me. They’re what I’m now doing differently.
- Structure the scene before you prompt. The workshop uses a set of five questions to make sure the scene is fully thought through before you write the prompt.
- Use style transfer. Point the AI to a design reference and have it apply that style to your own image.
- Check your prompt. A lot of the time the image goes wrong because of the prompt. Instructions like “make it more premium” or “make it less AI” don’t work. Describe exactly what you want, or point the AI to a reference.
Prompting well still plays a huge role. For me, the bigger shift was learning to see what’s wrong.