Pionyrka F1
The first version. Seems to be working well at strenght 0.7 - 1.0, no triggers needed, just use generic words like shirt, skirt, neckerchief and specify colors or other detailed attributes. During training was also specified sequence "Czech schoolgirl uniform", so it makes some impact too. Pictures can be generated at 2Mpx directly without any upscaling. Compatibility with other LoRas is surprisingly good, much better than in SDXL (tested several dozens of celebrities/faces). Tested on DEV fp16 only and to fit in 24GB VRAM with some other LoRa it requires to be merged first (see the merged version with details).