My first go making LoRAs here on Civitai - used the default settings, this is after 10 Epochs.
Seems to work a lot better than my TI of her, but still not great.
Three further attempts to refine this LoRA have not resulted in an improvement I'm afraid, even using entirely different training data.
I'm therefore just going to have to leave this one to stand as is.
Please be gentle with any ratings, I know it's a long way from perfect, but it may be of use to someone!
Recommended to use LoRA weight of 0.9-1.0, relatively low CFG (~5) seems best.
I'm still not 100% happy - truly sharp images are elusive, you may have better results than me. Also, due to pretty much every source image of her being with her trademark grin it's almost impossible to get rid of.
描述:
v1.0
Still needs some work, I will attempt v2.0 when I have enough Buzz again.
训练词语: NatVale
名称: 201365_training_data.zip
大小 (KB): 9258
类型: Training Data
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
名称: Natasha_Vale.safetensors
大小 (KB): 36983
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
