<h2 id="heading-13">RM02a: Experimental Base Merge</h2>I will update the previously published reference model RM01. After investigating in more detail what changes were made due to licensing issues and how licenses are handled during use, we found that some things were difficult to handle, so we will define a new reference model. There will be no changes to existing models.This time, at CIVITAI,No selling modelNo generation servicesI have excluded models that have more than the license limit set. Similar to RM01, I merged 5 types based on the number of downloads. Broadly speaking, there are five types: 2,5D, Photo Asian, Photo European, Anime, and Cute Anime, and this is also the same as RM01.以前公開したリファレンスモデルRM01を更新します。ライセンス問題で変更があったものや，使用時のライセンスの扱いについてより詳しく調べたところ，扱いにくいものが判明したため，リファレンスモデルを新たに定義します。既存のモデルなどについてはとくに変更いたしません。今回は，CIVITAIで，No selling modelNo genertion servicesのライセンス制限以上のものが設定されているモデルは除外いたしました。RM01と同様にダウンロード数を目安に5種類をマージします。大きく分けて，2,5D，Photo Asian，Photo Europian，Anime，Cute Animeの5種類になっており，これもRM01と変わりません。<a rel="ugc" href="https://civarchive.com/models/4384?modelVersionId=128713">DreamShaper</a><a rel="ugc" href="https://civarchive.com/models/69158?modelVersionId=287641">Meichidark_Mix</a><a rel="ugc" href="https://civarchive.com/models/15003?modelVersionId=372799">CyberRealistic</a><a rel="ugc" href="https://civarchive.com/models/2583?modelVersionId=106922">Hassaku</a><a rel="ugc" href="https://civarchive.com/models/121696/rabbit">Rabbit</a>I merged the above 5 types evenly.This time I used the latest version of each, but I felt that it was a little too learned as a base model, so I tried reverting the v15 basic model by 10%.I mixed a lot of things together, so the basics are a little more like a painting than a photograph. The faces I see are somewhere between Asian and Western, so I guess things are going well. If you add something like ((realistic)), it will look pretty good as is.　Any changes to the licenses of these models from now on should not affect the license of this model, so I think this is almost the final version. It may be possible to create a standard model with a license that can be generated by another company or sold.以上の5種類を均等にマージします。今回はそれぞれの最新版を使いましたが，ベースモデルとしては少し学習されすぎているように感じたので，v15の基礎モデルを10％戻してみました。いろいろ混ぜましたので，基本はやや写真寄りの絵画調です。アジア系と西洋系の中間くらいの顔が出てくるので，だいたいうまくいっているのでしょう。((realistic))くらいをつければ，そのままでも結構よい感じなります。以降，これらのモデルにライセンスの変更があっても，このモデルのライセンスには波及しないはずですので，ほぼこれで最終版になると思います。もしかしたら他社生成サービス可能ないし販売可能なライセンスでの基準モデル作成はあるかもしれませんが。By using reference models, creating merge models becomes extremely efficient. Even without using a reference model, you can achieve the same effect by using Lora-enabled ckpt based on a reference model. ckpt Lora can also be used as Style Lora, streamlining AI image production in many aspects.リファレンスモデルを使用することで，マージモデルの制作は非常に効率化されます。リファレンスモデルを使わなくても，リファレンスモデルベースでLora化されたckptを使うことで，同様の効果を得ることができます。ckpt LoraはスタイルLoraとしても利用でき，多くの局面でAI画像制作を効率化します。<h3 id="heading-4">RM01</h3>This is a model merely merged famous 5 models by same weight.<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/4384?modelVersionId=109123">DreamShaper 7</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/25694?modelVersionId=105035">epiCRealism 3</a> (c)epinikion</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/25494?modelVersionId=63786">Beautiful Realistic Asians 5</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/4468?modelVersionId=7425">Counterfeit 2.5</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/28169?modelVersionId=106439">CuteYukiMix EchoDimention</a></li></ul>4 most downloaded trained ckpts and defferent style anime model (their old models).So to speak, RM01 is a replacement for v1-5-pruned-emaonly.safetensors and is not intended to be used alone.Please read <a target="_blank" rel="ugc" href="https://civarchive.com/articles/3411/">my articles </a>for more information.Don't expect to get the best image by this, but I don't say this can't generate good images.

Files with SHA256: 0f270bb385cedaa96bd5e3323cd176dff52789ca63e1800b4d0546758667428a