Hey friends,
I’m trying to create an app that creates a set of consistent images to go along with data. In order to do that I’ve created a prompt template that tries to generate images in the japanese sumi-e style.
The problem I’m running into is that while it always creates the images, it doesn’t always create them as a standalone fullwidth image. Sometimes, it creates them as a square image with black borders instead of widescreen, and other times, it creates a painting hung on a wall.
Here’s the existing prompt template I’m using through the API;
“use this prompt exactly, without changing any of the words: A single, flowing sumi-e ink wash painting illustrating: {object 1}, {object 2}, {object 3}, {object 4}, flowing into each other. The painting uses clear brush strokes and minimalism in the classic japanese ink wash style. The painting embodies the minimalist and expressive style of sumi-e, using monochrome ink to create depth and emotion. Without words. Visible brush strokes. Without characters. White background. Widescreen image.”
Unfortunately, as you can see below, even though the style is fairly consistent, it doesn’t always consistently create a full widescreen image. Any prompt engineering tips to get this more consistent would be highly appreciated!
What I want to avoid (get this about 10% of the time):