Human/\pyramid

Lora was intended to draw people lined up in a pyramid formation for 6 people. The main dataset turns off exactly such drawings. But it does not consistently display more than three different characters. Large pyramids for 10+ people are especially bad. But these are already limitations of illustrious and sdxl models. They cannot display a large number of characters at the same time in detail.

Main trigger:

human/\pyramid
all fours, 3girls, 4girls, 6girls, 6+girls, 3boys, 4boys, 6+boys, so many

Options:

standing, trio, clone, carrying, multiple boys, multiple girls, chibi

Notes:

1. The number of characters will not always be the ones you wrote. "3girls", "3boys", -> often turn into "4girls", "4boys". With a large number of characters, you can add tags "6+girls" / "6+boys", "multiple girls", "multiple boys", "so many". But the quality of displaying a large number of characters is very poor (although the formation becomes more like a pyramid)

2. Because the keyword includes "Pyiramid" if the context and background are not specified, the background may show pyramids and ancient Egypt. I could make some neutral word that says nothing, for example "h2n3mp$r01mi5dDd" but I like the /\ option, it's easier to remember. Also, there are several images on danbooru with the tag human_pyramid and the first version of this lora was based on this tag.

3. The idea for this lora came while working on the lora for "Ninin ga Shinobuden". The ninjas from this anime sometimes built huge pyramids for people to show off their coolness. But not a single picture of ninja pyramids made it into the dataset for this lora. ¯\_(ツ)_/¯

4. Lora, as always, turned out to be not ideal. And the chances of a successful image depending on the number of characters are catastrophically reduced 3-4 ~ 50%, 6 ~ 30-40%, 6+ ~ 10-20%, 10+ ~ 0%