I'll be using comfy UI in the video but you could apply this to anything since most of the things I'm going to be doing is just focused on the prompt I also have a video about sdxl where I talk about creating photo realistic images with sdxl that video uses some extra tools and stuff in flux I won't be using any extra tools well partly because they just don't exist so that's a good enough reason but two because the prompt adherence in flux is much better so you can get more out of just using the prompt

we're going to swap the prompt from being a portrait of a woman to instead we have a portrait of a woman so the same thing and then we have subject the woman is wearing a t-shirt foreground a table background a door generally when you're generating images one way to make images look more realistic is to have more elements in your image

one of the big problems you're going to have with your prompt is that people have very smooth or plasticky look looking type skin that's one of the things you want to get away from and the image that we just had is very obviously smooth and plasticky type skin right so one of the ways you can try to get away from that is by communicating the sort of skin that you want

We have an image that we end up with, and she's clearly leaning over. She's standing up... We want to add her pants as well to the prompt because if we don't, it's going to automatically fill in something and we say she's wearing a pink blouse and khaki pants.

When you reduce the guidance, it tends to increase the realism of the image, but there's a tradeoff. The image itself suffers in terms of overall quality

We want to upscale the image. What it does is basically take patches of the image and upscale them individually. We can do a second pass, like a second pass is you take the image from the first sampler and you upscale it a little bit and you run it again through the denoiser