Trained on the animefull-final model for 12 epochs at a network dim of 16, then resized to one of 8.
Most of the images in the dataset were character portraits taken from the game with a white background, backgrounds still generate fine but they seem to weaken the style a bit.
The shading style seems to be best preserved with the Anything V3 model, as it keeps the flat coloring the best, the Based64 and 7th Anime models are not as accurate but still give very nice looking results.
The LoRA is trained to work at a weight of 1 but you're free to adjust it depending on what model you're using it on and what results you want.
There's no activation tag, just load the LoRA in and it should work right away.
描述:
训练词语:
名称: Shantae and the Pirate's Curse (Style).safetensors
大小 (KB): 10369
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
