Version 2 revamps the dataset entirely. She may almost look like a sibling compared to the previous version, instead of the same character. Now there is more detail to the face and skin, especially on closeup images. I noticed version 1 had facial features exaggerated from one image in my dataset that had a wide angle lens effect, so I made sure none of the new dataset worked that way. Same approach to her face and body features, but instead generated this version's data set in DPM++ 3M SDE Karras, used a variety of more detail enhancing LoRAs, and closeup images
Version 2 revamps the dataset entirely. She may almost look like a sibling compared to the previous version, instead of the same character. Now there is more detail to the face and skin, especially on closeup images. I noticed version 1 had facial features exaggerated from one image in my dataset that had a wide angle lens effect, so I made sure none of the new dataset worked that way. Same approach to her face and body features, but instead generated this version's data set in DPM++ 3M SDE Karras, used a variety of more detail enhancing LoRAs, and closeup images