(7/6)Uploaded chilled_remix version. Additional learning was done with illustrations generated by niji-journey to create LoRA. The trigger word is "jis". It is based on various models. Files are different for each base model. niji2：Own merge model zzzmix(https://huggingface.co/zzzAI19/zzzmix) niji2animekawa：AnimeKawa (https://civitai.com/models/87661/animekawa?modelVersionId=93295) niji2anything：Anything v5 (https://civitai.com/models/9409/or-anything-v5ink) niji2beautifulRealistic：Beautiful Realistic Asians (https://civitai.com/models/25494/brabeautiful-realistic-asians-v2) niji2chilooutmix：chilloutmix (https://civitai.com/models/6424/chilloutmix) niji2counterfeit：counterfeit v3 (https://huggingface.co/gsdf/Counterfeit-V3.0) niji2sukumizumix：SukumizuMix (https://huggingface.co/AkariH/SukumizuMix) niji2chilledremix、niji2chilledreversemix：chilled_remix(https://huggingface.co/sazyou-roukaku/chilled_remix)
I also plan to use TrinArt, Irismix, and openjourney as base models in the future. LoRA with these models as base models will be uploaded tomorrow. I would recommend a LoRA strength of 0.7. We also consider step 6 to be a good choice. Sample images can be found at https://ai-drawing.net/en/2023/07/05/introduction-of-niji-lora-v2-0/
(7/6)chilled_remix版をアップロードしました。 niji・journeyにより生成されたイラストで追加学習し、LoRAを作りました。トリガーワードは「jis」です。 色々なモデルをベースにしています。ベースモデルごとにファイルが異なります。 niji2：自作マージモデルzzzmix(https://huggingface.co/zzzAI19/zzzmix) niji2animekawa：AnimeKawa (https://civitai.com/models/87661/animekawa?modelVersionId=93295) niji2anything：Anything v5 (https://civitai.com/models/9409/or-anything-v5ink) niji2beautifulRealistic：Beautiful Realistic Asians (https://civitai.com/models/25494/brabeautiful-realistic-asians-v2) niji2chilooutmix：chilloutmix (https://civitai.com/models/6424/chilloutmix) niji2counterfeit：counterfeit v3 (https://huggingface.co/gsdf/Counterfeit-V3.0) niji2sukumizumix：SukumizuMix (https://huggingface.co/AkariH/SukumizuMix) niji2chilledremix、niji2chilledreversemix：chilled_remix(https://huggingface.co/sazyou-roukaku/chilled_remix)
また今後、TrinArt、Irismix、openjourneyもベースモデルにする予定です。 これらのモデルをベースモデルとしたLoRAは明日アップロードする予定です。
LoRA強度は0.7を推奨します。 またステップ6が良いと考えます。 サンプル画像は以下のページにあります。 https://ai-drawing.net/2023/07/05/niji-lora-v2-0%e3%81%ae%e7%b4%b9%e4%bb%8b/