🧩 版本说明 — v1.5（实验版）

📘 概述（Overview）

日语：
本模型以 Illustrious v1.0 为基础，大幅修改了标注方式，属于实验性版本。
训练数据本身并未做重大改动，而是旨在通过重新审视标注方法，验证其对输出倾向的影响。
此外，本次整合了我日常使用的VAE，默认状态下可还原更自然的色彩表现。
如需更换VAE，可轻松替换为您自己的VAE。
由于标注方式的变更，我感觉模型对LoRA和Prompt的权重（weight）反应更加敏感。
不过，初步测试中发现眼睛形状容易出现轻微变形，未来版本中将对此进行调整。
因此，v1.0中存在的一些弱点可能仍部分保留，但模型对调整的响应性已有所提升。

英语：
This experimental version (v1.5) is based on Illustrious v1.0 and introduces a major change in annotation method.
The dataset itself remains mostly the same, but the goal is to evaluate how annotation differences influence generation results.
Additionally, I have integrated the VAE I personally use, providing more natural color reproduction by default.
You can replace it with your preferred VAE if desired.
Due to the new annotation approach, the model now responds more sensitively to prompt and LoRA weight adjustments.
However, from early testing, I’ve noticed that the eyes tend to deform slightly more easily in some generations — something I plan to refine in future updates.
While some weaknesses from v1.0 may still persist, the model’s responsiveness and adaptability have clearly improved.

🧠 开发意图（Purpose）

日语：
为在正式发布v2.0之前，在中间阶段验证标注方法的影响而推出的实验版本。
在保留原Illustrious特色的同时，整体氛围给人以更接近Kadenz系列的印象。
本次整合VAE后，饱和度与明暗表现更加稳定，对权重调整的响应性也得到提升。
不过，目前仍存在眼部形状易崩坏的情况，将成为后续学习与修正的重点。

英语：
This version acts as an intermediate experimental checkpoint before the full v2.0 release, created to analyze the effects of new annotation methods.
It maintains the Illustrious identity while leaning slightly toward the Kadenz series in tone and balance.
The integrated VAE improves tonal stability, and the model now reacts more effectively to weight changes.
That said, there are still occasional cases where eye structure becomes unstable, which will be one of the focus points for the next update.

💬 请求反馈（Feedback）

日语：
由于标注方式变更、VAE整合、权重响应性提升等多重要素叠加，
恳请提供实际使用体验与对比反馈。
例如：“v1.0中存在〇〇问题，而v1.5中〇〇得到改善”等评论将特别有参考价值。
关于眼睛变形程度、与LoRA的兼容性等，欢迎提出广泛意见。
希望能与大家共同验证标注变更的“威力”及权重调整的效果。

英语：
Feedback is highly welcome — especially direct comparisons such as

“v1.0 had X issue but v1.5 improved it,”
“eye rendering still needs tuning,” or “LoRA weights respond more clearly.”
I’m especially interested in how users experience the new annotation system, integrated VAE, and weight responsiveness in practice.
Your insights will help refine both visual stability and realism for v2.0.

🪶 感谢您测试此实验版本。您的观察与反馈将直接塑造即将推出的v2.0。