FYI I'm using cthulhu11's training method for generating Flux LORAs on CivitAI's on-site trainer. He seems to be gone for a while and since his FAQ seems to be gone as well, I'll leave it here if you're interested: First, gather 60 high-quality images for training (I recommend cropping out any watermarks). Then, use Auto Label to generate captions, but go through manually to make sure everything looks correct. Set your training parameters to Resolution 1024 (which should default the Train Batch Size to 2), roughly 2100 steps (usually 10 epochs and 7 repeats), clip skip 2, and optimizer Prodigy. Leave the rest at default settings. Finally make sure you do NOT leave sample prompts empty, you need these to test between epochs. And then presto, no one will see your hard work because CivitAI are being dorks. Get bent, losers!