Welcome to the Selfie Support Lora.
There isn't much to say, really: The goal of this Lora is to mimic natural selfie poses that look more "realistic." By natural selfies, I don't mean mirror selfies (Flux already handles those well) but just regular ones.
Training details:
180 HQ selfie images
AI Toolkit
The size of the Lora is slightly larger because I used a different LR and Alpha channels (64)
Trigger word: Selfie
Important: Phrases like "a woman taking a selfie" often resulted in mirror selfies. Simply using "Selfie, a woman in a bathroom" yields good results.
Sometimes, I've set the LoRA to 1.2 to "force" it into the desired selfie pose.
Might be useful for some of you! :)
