Mmh well what I like doing personnally, is I take an image as template
trace the general outline that i liked about the original character, tweak it to what I like, redo the shading / light angle and the pose, and essentialy a half scale sketch without doing a full drawing, either digitally or with paper,
I do 2-3 variations and feed it to ChatGPT, which I believe has the better "mass" image generation/editing capabilities (ofc LLMs in general aren't very ethical due to their training methods and data and it is better to use it sparingly, even better to host it yourself), then I see what it has done, and i tweak it digitally because im not always satisfied w results