Warriors at the Roman Baths
Roman Baths Series
This is an ongoing series that I’ve been working on since the very beginning. I like having multiple people in an image that creates dynamics and draw your eyes around the drawing / painting.
The most challenging aspect of doing this in AI is having all of the figures have strong expressions, correct number of fingers, and also logical compositions. AI engines excel at rendering the forms according to the artistic styles you have asked for, but where it fails is the logical composition of the structures. Since diffusion tech does not use 3D models with skeletons, it’s not uncommon to find limbs going to places where they don’t belong. It’s really easy to see it if you ask AIs to generate two men wrestling. I have done a lot of that and I never post them because it’s very hard to get a usable image. Try it.
Since Midjourney doesn’t have in-painting, I had to generate a lot of images in order to get something close. That’s also partially why I started using Stable Diffusion — in order to have much finer control for figures and limbs. What I haven’t been able to do so very well is to have the art be drawn in a manner that I like, and that’s why I experimented with style transfers. I hope that eventually my workflow can give me the best of both worlds.
Text prompts in Midjourney v5.