@hideki The good thing is that you can run StableDiffusion however you'd like, and that you have lots of people doing the prompt engineering for you because of this.
StableDiffusion is, I think, the worst in needing looong prompts to get cool shit. But it's worth it.
"A portrait of a samurai, young with long hair, wearing white robe, brandishing katana with red hilt, matte painting, hyper realistic, very detailed, intricate, forest, unreal engine, dramatic cinematic lighting rendered by octane, 8 k, detailed"
Steps: 50, Sampler: k-diffusion, CFG scale: 7, Seed: 3866088339, GFPGAN
using everything as described in
https://rentry.org/kretard
Same thing, exactly same prompt, and i prefer Dall-E results because they are more aligned to what i am looking for, despite being less detailed.