concept In my arms[Experimental]

这是一个实验品，英文部分我不会润色校正

This is an experimental product, I will not polish and correct the English part

首先我不得不说这个制作比较失败，它过拟合严重，肢体崩溃还很难用。所以我把它暂时定为一个实验品。如例图所示这个lora可以生成一个抱在怀里的第一视角，我很喜欢这个，但是我的能力暂时没法把它做更好。我的使用办法是先批量生成一些图片，分辨率大概是900x600，我使用的权重是1，你也可以自己调整。这些小图不必管画风是否过拟，只要求肢体正确，挑取构图和肢体比较合适的一张后，再使用图生图和局部重绘进行画风和细节修正。如此复杂才能做出一张差强人意的图，如果你并不是特别追求这个就别试了。

First of all, I have to say that this production is a failure. It overfits seriously, and the limbs are broken and it is difficult to use. So I set it up as an experiment for the time being. As shown in the example picture, this lora can generate a first-person perspective in the arms. I like this very much, but I can't make it better for the time being. The way I use it is to generate some pictures in batches first, the resolution is about 900x600, and the weight I use is 1, you can also adjust it yourself. These small pictures don’t need to care about whether the painting style is over-drawn, but only require the body to be correct. After picking a picture with a more suitable composition and body, use the picture-generated picture and partial redrawing to correct the style and details.It's so complicated to make a passable picture, don't try it if you don't particularly pursue this.

我例图的prompt：in my arms, 1girl,1boy,pov,grabbing,grabbing from behind, breast grab,from above, mixed bathing, water,completely nude, nipples

我建议多使用”mixed bathing, water“来掩盖躯体变形

I suggest using more "mixed bathing, water" to cover up body deformation

以我例图第一张为例，我的步骤参数大概如下（具体是多少不重要）：

Taking the first picture of my example as an example, my step parameters are roughly as follows (the exact number is not important):

草图：分辨率896x640 采样器euler a 步数20 lora权重1

Sketch: resolution 896x640 sampler euler a steps 20 lora weight 1

局部重绘：分辨率不变 幅度0.65 去除了多余的手并简单做了一些修正

inpaint: same resolution, amplitude 0.65, removed redundant hands and made some simple corrections

图生图：分辨率1344x960 幅度0.35 不使用lora 做了放大处理和画风修正

i2i: resolution 1344x960, amplitude 0.35, without using lora to do zoom-in processing and painting style correction