Prone hone (multiple views) (3koma)

强度：1-1.3

这个lora主要作用是展示prone hone动作中的前后侧三视图，仅固定了姿势和视角，其余要素可自由调整。

使用方法：

• 最推荐使用正方形画幅，或者4：5的宽高比，更细长的比例下难以保持固定的构图。

• 完整复制第一段触发词，尽量不要改动，然后再额外添加需要的外貌、动作、场景提示词即可。如果需要展示脚部，则添加“legs up”或“feet up”。（可以参考预览图）

已知问题：

• 最严重的问题是画质劣化。我不知道是因为我的素材太少（225）、画风太单一，还是因为每个视角占画面的像素太少了，或许是两者都有。总之这个lora直接出图会降低画质，推荐使用二次采样，一采使用lora来固定构图，二采去掉lora来重绘画面。

• 即使添加了内裤的提示词，在后视图中仍无法显示，这个确实是我的素材问题。

• 构图错误和人体扭曲出现的概率并不高但仍然存在，请多抽卡。

---

Strength: 1-1.3

This Lora is designed to show front, back, and side views of the prone hone pose. It only fixes the pose and viewing angle, while other elements can be freely customized.

How to use:

• The most recommended aspect ratios are square or 4:5. More elongated ratios may make it difficult to maintain the intended composition.

• Copy the first part of the prompt exactly — try not to change it. Then, add extra prompts for appearance, action, or background as needed. To show the feet, add "legs up" or "feet up". (Suggest referring to preview image)

Known issues:

• The most serious issue is a drop in image quality. It's unclear whether this is due to too few training images (225), limited art style variety, or because each view occupies too few pixels — possibly a mix of both.For best results, it's recommended to use a two sampling process: first use the Lora to fix the composition, then remove it for a second sampling pass to refine the image.

• Even when adding prompts for panties, it not appear in the back view. This is likely due to limitations in the training data.

• Composition errors and body distortions are uncommon but still possible.Generating multiple times is also advised.