I knew little to nothing about making textual inversions when I made my Panda one, so keeping with tradition, for this LORA I honestly just downloaded a software, fed it the same dataset as the TI, clicked a few things and hoped the default settings would work. They did, I think. Anyway, try keeping it around 60%, it seems to give the most… desirable results.
Also it seems no people allowed? I guess my dataset being so clean with no people in shot kinda caused it. Oh well. Never say no to Panda! It will bring glory to your house!
There's a few versions. The original, that is ok. version 2, that is improved. 2XL, that is based on SDXL and IMO not great. Then 2AV that is based on AniVerse 4 https://civitai.com/models/107842?modelVersionId=438916
描述:
I trained the same exact dataset on SDXL! That said, I think it's better to do everything with the 1.5 ones....
训练词语: fiat_panda
名称: fiat_panda2xl_2000_lora_f32.safetensors
大小 (KB): 479967
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
