1
How to prevent "spilling description" of one element into another (lemmy.world)
submitted 1 year ago by to c/perchance@lemmy.world
14 comments fedilink hide all child comments
 

When I am describing a scene in the prompt of https://perchance.org/ai-text-to-image-generator for several elements lets say three persons, I want to describe their characteristics and give them some features for example three women, first women has blonde hair, glasses and neck tattoo, second woman is brunette wearing pink hat and smoking and third woman is redhead with long hair, eyebrow piercing and sticks her tongue out What AI does is to mix this description between all the characters, I tried to add thing like does not smoke, does no wear glasses ect. to description of each character but that didn't have much effect. Any idea how to "border" description for each element?

you are viewing a single comment's thread
view the rest of the comments
[–] [S] 1 point 1 year ago (2 children)

Thanks I will try to play with it, are these BREAK, parenthesis () colon : semi-colon ; and dot . some universal keywords and characters or that is just a thing that seems to work with perchance?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 2 points 1 year ago (1 child)

    They are (seem to be) universal

    • I know BREAK is and works like advertised.
    • The parentheses can be used to add more weight to a (piece) of a prompt. Given that no parentheses is a weight of 1, then a single () is 1.1 * 1 weight, (()) is 1.1 * 1.1 * 1 weight and so on. Or they can be used like: (some really important piece of prompt):1.21, adding a certain weight directly. Don't overdo the decimal, something between 1.1 and 1.7 will usually do the trick.
    • The colon : is often used like a sum-up, e.g. the female is wearing: ..., ..., ... and ... ending with a dot or a semi-colon.
    • The semi-colon ; is used to split part of a scene like they are dancing; optimal lighting, low camera angle; leisure time concept.

    It's not an exact sciene, but it will help.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] [S] 1 point 1 year ago

    Interesting is that this works quite well in perchance but if I try to put what you wrote in prompt in comfyUI with standard text-to-image workflow regardless of a base mode used (SD 1.5, SD 3, SDXL 1, Pony, Ilustrious) result is an absolute mess, it generates on average from 2 to 5 characters, dresses, prompts and races are utterly random, I wonder what magic is used in perchance that results are so much accurate to prompt

  • permalink
  • fedilink
  • source
  • parent