Try size: 768 x 1024 or 1024 x 768 or 1024 x 1024 Weights: 0.5~1.0Clip skip: 2Training material from web pics and bananaNote: If overfitting occurs, reduce the weights to improve the situation. / 如果出現過擬合的情形，請降低權重來改善狀況。(Anima10)Note: I've wanted to train Jiufen LoRa since last year, but I wasn't confident about the materials and parameters, so I kept putting it off until now. The scenes is quite complex, so I suggest referring to the prompt in the training set. I don't recommend generating realistic characters; my plan is to focus on anime characters. / 從去年就想訓練九份LoRa了，但那時對素材跟參數沒有把握，所以一直拖到現在。場景比較複雜，所以建議參考訓練集中的完整提詞。不建議生成寫實人物，我的計劃就是針對動漫人物。(ILXL)

jiufen / ruifang / (九份 / 瑞芳)

jiufenoldstreet_zimagedeturbo.safetensors

Files with SHA256: e664d330f3b56efffbd9b205dd92d556d9062c9fad159a74ec84549610b7f681