I'd be interested to know how your model performs if it is trained only on your own work.

Assuming you haven't taken a photo a second for your entire life, then I suspect you'll struggle to make something even close to what's available publically, due to lack of training data.