Trained in 3 diferrent datasets consisting 28-40 imgs each.

The 1st one is focusing on 2021-2022 images and full body. Works well with any Magmix checkpoints.

The 2nd one is more diverse 2018-2024 images, with more face focus. Works well with FluxDev with more variations.

The V3 has two version: younger, before 2015-2017 datasets; older, 2023-2026 datasets.

Just give it a try.