so as shared in a previous post i started trying to learn about training custom models on top of Stable Diffusion, by training it to know how to draw Raekwon of Wu Tang.
the process is still an ongoing exploration, but i do have some comparisons of my custom model compared to the base model with txt2img using same parameters and prompt, and seed set to -1 (random)..
here’s the base model’s results
the results of the base model, helped me see why i was seeing similar results as there’s a common appearance that repeats, which i see in some of my own responses. the results are OK but definitely couldn’t pass as Raekwon.
ok here’s my model that’s been trained to do better…
WOW! i can say with confidence there’s definitely an improvement.