![ufoTable artstyle版本LoRA [Experimental] (ID: 124455)](https://image.1111down.com/xG1nkqKTMzGDvpLrqFT7WA/d701e4ae-12bc-488c-9718-1434781c1a9a/width=450/1694600.jpeg)
A ufotable screencap, i recommend you to use the LoHA if you wanted better results
I haven't tested much the LoRA version so you'll have to find out yourself how to use it, and honestly i don't know what's the suitable model for it. Use ESRGAN or latent upscaler with it, it was A test on training on 896x896 and a new config
I recommend you to use the LoHA
Recommended weight for the LoCon and the LoHA: 0.3~0.45 if you didn't fate artstyle to take over the character you want, keep it in your mind that most of Dataset was gathered from fate, i recommend 0.3 the most. It works good, for upscaling use ultrasharp-4x with denoising 0.3 & Tiling [Multidifussion] Method
0.5 and above if you want to do a chara in fate style
描述:
A training test on 896x896 and a new config
训练词语:
名称: ufu.safetensors
大小 (KB): 36976
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success