Introduction:

v1:

This is an initial experimental version. It’s an interesting attempt, and I hope everyone can give this fine-tuning a try as well.

v1 was created using a training set of 428 specialized images focused on scene enhancement and adjusting the character-to-background ratio.

This results in a more balanced composition between characters and scenes, as well as heightened sensitivity and lavishness in scene-related prompts.

Of course, if the prompt emphasizes character descriptions, the model will still display a unique stylistic flair and is fully compatible with various artist-related prompts.

The usage approach is similar to Illustrious-XL-v0.1 and NoobAI-XL (NAI-XL) Epsilon-pred 1.0-Version:

<1girl/1boy/1other/...>, <character>, <series>, <artists>, <special tags>, <general tags>

Base Model Used for Training:

v1:

NoobAI-XL (NAI-XL) Epsilon-pred 1.0-Version

Training Process:

v1:

Used: LyCORIS

• Algorithm: LoKr

• Optimizer: Lion

The final model was created by merging the trained LoKr with the base model, NoobAI-XL 1.0, at a weight of 1.0. It was not merged with any other models.

簡介

v1：

這是初步實驗的版本，是個有趣的嘗試，歡迎大家也可以試試這個微調。

v1使用了428張特化場景和人物與場景比例調整的訓練集，讓場景和人物占比更加平均，以及提升場景相關題詞的敏感度和華麗程度。

當然，如果提詞主要還是描述人物的話，依舊有風格上的變化，且依舊能和各種藝術家提詞組合搭配。

使用方式基本和Illustrious-XL-v0.1以及NoobAI-XL (NAI-XL) Epsilon-pred 1.0-Version相同。

<1girl/1boy/1other/...>, <character>, <series>, <artists>, <special tags>, <general tags>

訓練底模：

v1：

NoobAI-XL (NAI-XL) Epsilon-pred 1.0-Version

訓練流程：

v1：

使用LyCORIS

• Algorithm: LoKr

• Optimizer: Lion

最終模型是以底模，也就是NoobAI-XL 1.0合併經訓練的Lokr，以權重1.0合併而成，沒有和其他模型合併。