Want Better AI Images? Ask a Chatbot!
How to use LLMs to brainstorm scene ideas, remix pictures, and more.
Happy Thursday, digital dingos!
Almost exactly1 a year ago, I wrote about “splatterprompting.”
That’s the practice of filling text-to-image prompts with dubious descriptors like “64K, HD, hyperrealistic, award-winning, masterpiece, best, epic, legendary, godlike, Captain-Marvelly”…you get the picture.
The gist of my argument was: “Stop that.”
My sentiment still holds.
Today, in line with the “connecting the dots” pledge I made last week, I want to demystify text-to-image prompting for those who still think it’s a complex and incomprehensible science.
Follow me, and you’ll see just how simple it can be…2
Image models have gotten way better
In my now-one-year-old article, I made the case for using natural language instead of splatterprompting:


