As far as I know, he uses the Novel AI model, and he certainly put a lot of effort into crafting his prompts. Perhaps you could try using more camera-control prompts to achieve the effect you're looking for? If that doesn't work, you might want to try some of the excellent LoRAs designed for camera control; I think that should get you closer to the desired result. I didn't focus on camera angles or composition control when tagging the data—that is a limitation of the local model.