To improve the user experience, would it be feasible to combine the --diffusion-model and --model flags by having the application perform an initial examination of the model file to see what it needs and adapt to supplied flags or error with a useful message?