Each of these embeddings produces a different more-or-less consistent synthetic woman. These embeddings were used to produce images that were then cherry-picked for trainings data to train LoRAs. Because of the cherry picking, the LoRAs may produce more consistent characters. Using the embedding and LoRA together during generation might help improve the consistency of generation on SDXL. See the trained LoRAs at
https://civitai.com/models/489555 (for sd15)
https://civitai.com/models/483639 (for SDXL)
https://civitai.com/models/487836 (for Pony)
See "About this version" for each character's bio and dating profile.