My own attempt at product photos was a total mess (bad lighting, weird shadows). The camera roll was just twenty versions of the same flat, head-on shot. It finally clicked that I hadn't given the viewer, or an AI, any real sense of the object's shape. A proper mix of angles makes all the difference, like showing the back of a bookshelf or the underside of a mug. You don't see the depth until you move around it.
Wait, that actually worked? It's wild that the fix was just adding more pictures from different sides. I guess the AI really did need that basic training like a little kid. Makes total sense now that you say it.