Duetothein*********ofLycoriscausedbysensitivitytolearningratesettings,theupdatedmodel(5.2version)isstillconventionalLoRAmodel.Useimagespublishedbysomeusersasdatasettotrainthemodel.Fixthecompatibilityproblemofnon-NAI-basedcheckpoints.NowarbitraryanimemodelwithNAI'sVAEorkl-f8-anime2VAEcanalsogenerategoodresultsusingthisLoRA,theoretically.Thisversionisabitoverfittedthatwillbefixednexttime.