That's not Stable Diffusion though, it's a different third-party model. (MiDaS, for example, but there are many of those) Generally, good depth maps is a holy

That's not Stable Diffusion though, it's a different third-party model. (MiDaS, for example, but there are many of those) Generally, good depth maps is a holy grail of imaging, but the issue is that it's very sensitive to resolution and correctness, so it must be perfect and have better resolution than the picture to achieve good results. ML-powered depth estimation isn't there yet.