My wife scolded me for not going to bed early last night. I said, “Just one more try!”
No, I was not doing anything naughty or playing Nioh 2 (an amazing Japanese action game). I was generating images in Flux, the new open-source (aka FREE) text-to-image Generative AI engine.
My prompt:
“traditional oil painting of a young bald asian man with glasses, he is crafting red crystal birds, in a traditional woodcraft workshop. Light is streaming through the window on the side. A curious blue sparrow is perched at the window sill staring at the man.”
As you can see, Flux can understand and generate many elements of the prompt. However, it has inherent bias where it kept giving me an old Asian man instead of a young one. This implies that it was trained on only images of old bald Asian guys. C’mon, I’ve been bald since my 20s, don’t discriminate against young bald Asian guys!
Flux is free and runs on your PC, but the requirements are heavy. You need a powerful gaming GPU and lots of RAM (32GB) to run it well. Also, the interface is very complicated, and it takes many iterations to get an image right. I went through 20 iterations of this image, which is about 30min of experimentation.