<pre><code>Alas, I found that I was making a very wrong lora. I thought before that this might be eye-catching. When I drew the picture in 2.5D, I thought it was pretty good. It could perfectly reproduce the jumpsuit taken off to the waist, but after When tested on a realistic model, it's another story. It becomes separate sleeves and transparent pants, and the connection between the sleeves and pants disappears. Depressed.

But, having done it, I definitely have to share it with everyone so that everyone can be depressed with me, haha.

Let's get back to the subject, let's talk about the prompt words. This lora has two sets of prompt words, which are marked separately, trigger words added separately, and then put together for training. The weight of 0.8 is the best. If it is 0.9, it will be better in realism. Fitting, because I deducted faces separately from the training set, so when a white face appears, I can detect it immediately (if the entire head is deducted from the training set, there is no way to find out whether it is overfitting, and I can only rely on giving verbs The prompt word has a higher weight). In the prompt word for undressing, if the navel is not written, the jumpsuit will be kept below the chest. When the navel is added, the jumpsuit will be taken off to the pussy position. In fact, it is The difference between one zipper and zipper is that there is no word zipper when identifying the tag. The other one is a relatively complete one-piece suit. There is not much more to say. I hope you all have fun.</code></pre>

Files with SHA256: 329fe289978f71ddd91d554c43186e4977c2ca90863b75f2c25b39059dd47570