
I was highly offended that nobody had made a Flux lora of best girl yet so I just did it myself.
Version 3
V3 is out and is once again significantly better than the previous version. This will likely be the final version as I'm happy with where it is at. The most commonly used tags in the dataset and the ones I recommend using are:
hutao, chinese clothes, long sleeves, black shorts, porkpie hat, red flower
I also highly recommend using the flat color anime lora as it seems to fix some weird hand generation.
Version 2
V2 was trained on more than 30x as many images as V1, so it's understanding of Hu Tao's appearance is much better. Unfortunately using Danbooru tags for the dataset seems to cause weird prompting issues with Flux, so for V3 I will look into using more natural prompting.
Version 1
V1 has a very weak understanding of Hu Tao, I do not recommend using it. If you do then you need to use many descriptive tags as possible otherwise it doesn't really look like her. The ones I use are:
hutao, brown hair, long hair, twintails, red eyes, symbol-shaped pupils, black coat, black hat, hat flower
描述:
训练词语: hutao
名称: Hu_Tao_V3.safetensors
大小 (KB): 18848
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success