Trained on 768x1024 landscape images , with composites of 12-20 smaller photos of 'clothing averse' women focused on face and shoulders.
The keyword used is 'xxx photograph' at the start of every training prompt using CLIP Skip 1.
Note that the FLUX Schnell model is optimized for Euler 10 steps at 3.5 CFG using the 'simple' scheduler
This LoRa strongly favours blonde women.
It mixes well with the CGI-effect lora at a 0.3-0.6 mix rate: https://tensor.art/models/808151796989267683?source_id=njq1pFzjlEOwpPEpaXny-xcu
Images used are a mix of scaled down CGI-created renders and SD1.5 generations and SDXL generations.
Roughly 12 to 20 such images are placed within the 768x1024 landscape training image.
This LoRa works well at generating crowds and will usually minimize the size of individuals in images.