For best outputs: Sampler: DPM++ 2M Karras. Steps: 30, cfg 7, CLIP skip 1. VAE: none required. Pos prompt: the model is set to something between semi-realism to realism by default, but it can be pushed to very close to true photorealism, for example, wrap your text between "Photorealistic, extremely realistic, hyper realism" and "masterpiece, best quality, hyper detailed, ultra detailed, super realistic, realistic, real photo, realistic skin texture, sharp details, 8k, HDR, f.1/2, 70mm lens, Kodak, Analog style". Also, you may use the LoRa "HKStyle_V3" (found in some of my example runs) with a weight of 0.8 to gain some more visual quality and realism, being "hkstyle" its trigger. This LoRa is thought to create knights, so, keywords like "sword, armor" or "helmet" may be needed in the neg. Neg prompt: "text, watermark, low quality, worst quality, bad compostion, poor, low effort, child, childish, young, nude, nudity". Checkpoint Trained.