This is a ComfyUI workflow using only components available in the Manager. It does:<ul><li>A DynamicPrompts txt2img using Juggernaut SDXL and a LoRA - output 1 Megapixel size, 4X5 portrait. </li><li>It then does face detection for a SINGLE FACE with a reasonable amount of feathering/cropping around the discovered face (you need model 'face_yolov8m.pt'; It shows you what was detected, so you can trouble-shoot).</li><li>It then does ADetailer inpainting of that face using the same LoRA and same weight, but a different prompt. This improves the face a lot if the face is distant, and allows you to be very face-specific in the inpainting prompt. </li><li>Finally, it does an upscale using UltraSharp.</li></ul>Advice:<ul><li>To take advantage of Dynamic Prompts, press "Run" a bunch of times in succession. You CANNOT get different prompts by batching (not by batching the Run, nor by batching the latent images!)</li><li>This runs happily on an RTX3060, with each image taking about a minute to generate</li></ul>

Files with SHA256: 51f858cbf71e6012995ba30c9fa6c1a0b75a7bce5f79f3be6949c2f6382acd2f