Trained on ~100 pictures I shot myself with Google Pixel phones over the last 2 years.Pair with <a rel="ugc" href="https://civarchive.com/models/1662740/lenovo-ultrareal?modelVersionId=2882170">Lenovo UltraReal</a> / <a rel="ugc" href="https://civarchive.com/models/1862761/nicegirls-ultrareal?modelVersionId=2882216">NiceGirls UltraReal</a> for (potentially) better results. I also highly recommend doing a hires pass with flux2/zit to get better skin and details (which is NOT done in the preview to showcase what the LoRA does alone).Not entirely satisfied with where this landed. Skin and background details are not there yet. And trying to teach them really shows how much Qwen VAE sucks.Will do another version later when I have found a solution to the detailing problem (loss change, higher res training, skin detail images, ...), if there is any base model that might just need more high resolution training

STYLE > PixelPhoto

Training Settings: batch size 4, 1024px, ~100 images per epoch, optimizer is SinkSGD_adv

Files with SHA256: af95d6d629f4d928544d8fc4bb76ce77f6b8667589ff3deb13ae33000222384f