Flux LoRA model trained with 31 screen caps from the original Sailor Moon series in 640x480.

I annotated the dataset by hand, providing natural language descriptions in logical sections:

1. tags

   • lora tag sailor_moon

   • style tag classic 1990s anime

   • composition tag full body shot , portrait, close-up portrait, cowboy shot, upper body shot, hero shot, close-up side portrait

1. Outfit description

   • recurring keywords: sailor uniform, white bodice, white top, sailor collar, red chest bow, red boots, white gloves, pleated skirt, red necklace, …

2. Hair description

   • recurring keywords: distinctive style, twin-tails, buns, bangs, long blonde hair, ponytails, …

3. Facial expression description

4. Scene and/or background description

I don't have a local flux install so I cannot really test the best epoch. I also made the mistake to let ComfyUI generate 2 of the 3 test prompts. So I only had one image to compare versions.

This is why I publish multiple epochs for you to test if you wish.