Hotori from Neverness to Everness
Description
FAQ
Comments (4)
The Anima model has better support and powerful capabilities for natural language. It is recommended to use the following prompt:
Hotori, with her long black hair and warm smile, wears a delicate hair ornament and jewelry that accentuate her red and orange eyes. Her bare shoulders and collarbone are framed by an off-shoulder black kimono adorned with red flowers and a matching choker, complemented by tassels, a hair flower, and elegant earrings. She holds a hand fan gracefully, her outfit completed with traditional tabi, miko-style white thigh-highs, high heels, and clog sandals, blending classic Japanese clothing with modern charm.
It's better to use both tags and natural language, actually. I just prefer tags. But thanks for your advice!
@Twisowk I also noticed this problem. Using natural language tends to focus on the character's focus (with a pure white or no background), while using prompt tags results in better character accuracy (such as details of clothing, etc.).
@GreenReimu, it's not really a problem, it's about training itself. My dataset captions contains only danbooru tags, not natural language. That's why model understands tags better.