All the pieces have been generated and upscaled offline on a RTX3090 and custom ComfyUI workflows.

Some - such as the towering structures - reach 12K in height.

Invoke and Fooocus have been used for deep inpainting sessions.

Color grades, touchups and DAM were done in Photoshop and Capture One.

LoRAs of real people have been trained with AI-Toolkit and an H100.