How would you phrase it if I’m trying to generate a pic where a tall person is standing next to a small one towering over them

  • MagicShel@programming.dev
    51·
    2 months ago

    Caveat: I haven’t really been generating pictures in months. Maybe close to a year. It’s a fast-moving space so my information could be out of date.

    It’s really hard to give two characters in a picture different attributes—they tend to bleed over. Whatever you put in that text box describes the picture and not elements within it.

    One technique is to group the attributes, such as two people, first person tall giant towering, second person small short diminutive. This might work but you might have to regenerate several times and get lucky. The more attributes you try to specify, the more likely they’ll bleed over.

    You’re probably better off generating the first character by itself and then selecting an area to fill with a second character. I’ve not done much of this, and it’s tool dependent. If all you have is a prompt box, you’ll probably have to go with the first method.

    But there’s one last thing you can try. Describe what you want and get a result that’s kinda close to what you want. Then lock the seed and vary the prompt slightly to see if you can dial it in better from there. Replace tall with towering, or big, or massive. Sometimes it totally breaks the composition, but other times it lets you tweak something to get closer to your vision.

    Anyway specifying two subjects in a single text prompt can be very difficult, and it depends on how much you want to specify and how much you’re willing to be surprised.