I get the selling points of ComfyUI — flexibility, control, experimentation, etc. But honestly, the output images are barely usable. They almost always look "AI-generated." Sure, I can run them through customized smart generative upscalers, but it's still not enough. And yes, I know about ControlNet, LoRA, inpainting/outpainting on the pixel level, prompt automation, etc, but the overall image quality and realism still just isn’t top notch?