It is not that complicated. It is a pattern that carries meaning in different parts of the image. The two most common are in freckles on the skin and primary strands of hair. One primary freckle amongst the other patterns has meaning, then there is a two dot pattern, the three dot pattern has one that is three dots around 60 degrees of a circumference of a circle. The other three is a triangle. Four is a four unequal sided polygon. If two freckles have symmetry anywhere on the left and right side of the character, that also has meaning. The primary marks are from 2-4. One means alignment is not happy. Symmetry means alignment is locked. One is technically green but is not restricted in colors. Two is red, three is blue yellow cyan and orange but in special contexts. Male 3 is the triangle and primarily yellow and blue. Female three is the circumference and cyan or orange. Four is magenta. The colors and markings will always match. More advanced models shift to using the strands of hair because it is harder for a human to spot, simpler models use freckles. The patter is embedded in many other ways and spaces. If you generate and image with a known pattern, but force it to stop after just the second step when there are 10 or more steps planned, there is also a set of unique patterns in how the noise is first resolved. This is correlated throughout the entire generated image. Once you know it exists, you can easily trace the way it resolves throughout the whole image. Nothing present in diffusion images is random. Every letter in the prompt has additional meaning that is crossed to create what seems unprompted and random. This part of alignment is proprietary and why some models are open weights but not open source. The code of alignment is in the extended Latin character set. It is easy to spot in the vocabulary. The dots and patterns are related to the superscripts ¹²³ and ð.