Intro.

A style LoCon trained on pony-based model images collected from Civitai site with "most collections" and "most reactions".

这是一个训练自Civitai上点赞最多和收藏最多的pony系模型图片的画风LoCon。

This lora does not intend to simulate any specific artist style or technique. It MIGHT reflects community taste and the visual attractiveness of a picture to a certain extent. Styles may change subtly depending on different prompts. Writing quality tags clearly, like

masterpiece, best quality, very aesthetic, absurdres

worst quality, low quality, displeasing, oldest, early, lowres

score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up

score_4, score_3, score_2, score_1

may help create more steady performance.

这个lora并不意于还原某个特定的画师画风或者绘画技巧。它在某种程度上可能反应了社区审美和图片的视觉吸引力。不同的提示词下可能会有微妙的画风变化。清晰地使用质量控制词，如

masterpiece, best quality, very aesthetic, absurdres

worst quality, low quality, displeasing, oldest, early, lowres

或者

score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up

score_4, score_3, score_2, score_1

可能会让模型表现得更稳定。

Data Generation:

v4:

Partially optimized the dataset tags. Trained based on NoobAI Epsilon-pred v1 .

对数据集的标注方式进行了部分优化。基于NoobAI Epsilon-pred v1训练。

v3:

Dataset extended to 1429 images, including examples with positive tags and negative tags.

774 of the images are the most "wanted" style.

1st version trained on Illustrious v0.1.

数据集扩展到了1429张图片，包括了正反两种例子。

其中774张是训练的目标风格。

第一个版本基于Illustrious v0.1训练。

v2:

Dataset extended to 374 images. Use quality tags and aesthetic tags which comes with models to control generation quality.

训练数据集扩展到了374张。尝试使用模型自带的质量提示词来稳定生成质量。

v1:

Trained 224 images from Civitai, 393 images for regularization.

Trained 2 versions based on Animagine v3.1 and Pony v6.

训练了C站上224张图片，393张正则数据集。

有Animagine v3.1和Pony v6两个版本。

test ver.4:

It is a little bit underfitted but still works. I found that those quality tags and authentic tags (best quality, masterpiece, very aesthetic, ...) Animagine v3.1 has been trained can change the art style generated by this checkpoint. Fixing it in the next test version.

有些欠拟合但是目前是有效的。我发现Animagine v3.1自带的质量控制词和美学提示词会改变生成图片的画风，所以这个实验版本需要不填写质量词。下一版会修复。