My first Lora) I wanted to make it better, but it's really hard to find a good source images.
I recomend using weights somewhere between 0.6-1.
Tags
(!) FOR v5 if you can't make a good picture use this prompt:
(bent over:1.3), (bentover:1.3), (female on tiptoes:1.3), from side, full body, man behind woman, 1boy, standing sex, standing doggystyle, hetero
And add description of the scene and girl. It's working on every model I tried it to.
There are should be some differences when you'll use this words (you doesn't have to use it):
tiptoesFuck, from side, from front, from behind, bent, standing, penis
Also I recomend using:
sex, standing_doggystyle
And for the negatives:
Loli, dwarf, hug, 2girls, 2boys, threesome, yuri
There are lots of arm grab and hair pulling in the source images. But for some reason i couldn't been able to invoke them.
Versions differencies
v1 is generic version
v2 New verion, which is slightly better with limbs, charactrers bodies and position. (I haven't used upscale on the examples for this version)
v3 Just works better, but affects style much stronger. I don't know why, but it works bad in anything v4.5. You can use weight 0.8-1 for anything, and then use imgToimg for details. Worked out for me.
v4 Now it gives you more control but also it's much harder to trigger. HiresFix sometimes better. Sometimes not)
v5 Now it much more stable in depicting the pose itself, but not in every model. Realistic models are doable too. Also, I recomend using the prompt from sample images - works good.
I am done with that model, I think, I am happy with the results)
Enjoy)
You can try out this Lora online using Dazzle.ai
Feel free to ask any questions here in comments, or in my discord channel.
Also I am making games with AI arts. They will be free in the future, but if you want to participate in making or have an early access - you can support me on patreon)
Also my wet hair Lora was banned here, I dunno why. You can download the last version of it from my patreon for free.
Description
slightly better with limbs, charactrers bodies and position
FAQ
Comments (6)
这是我的第一个Lora。花费了很长时间才取得这个可悲的结果，但我想分享并听取您的进一步培训建议。 我使用动漫图片和玉米框架进行训练。基本想法是创造一个女孩站在脚尖上并从后面被插的姿势。 一般来说，除了姿势之外，我尝试描述图片中的所有内容。但出于某种原因，它非常不稳定。它往往会生成两个以上的人物，姿势奇怪。但几乎每次女性都是踮着脚尖的，这是一个胜利……我猜……有点 =） 我推荐使用权重在0.6-0.8之间。 使用以下单词时，应该会有一些不同（您不必使用它）： 踮着脚尖插入、从侧面、从正面、从后面、弯曲、站立、阴茎。 我还建议使用： 性交、站立狗仔式。 负面的包括： 萝莉、侏儒、拥抱、两个女孩、两个男孩、三人行、百合。 源图像中有许多抓臂和拉头发的动作。但由于某种原因，我无法调用它们。 v1是通用版本， v2是新版本，肢体、角色身体和位置都稍微好一些。（我没有在这个版本的示例中使用升级） v3效果更好，但风格影响更强。我不知道为什么，但它在任何v4.5中都表现不佳。您可以为任何东西使用0.8-1的权重，然后使用imgToimg来处理细节。对我而言行得通。 *仍在尝试使其按预期工作，所以我希望很快可以制作出更好的版本。 好的结果不稳定，您每10次或更少会得到2个好图像。所以需要一些耐心。ImgToImg也会随机修复一些奇怪的肢体连接。
BRO PLS MAKE REALISTIC VERSION OF IT!!!
@gigsa It's pretty flexible and working on realistic versions. Not fantasticly, but it works)
I tried to retrain it specifically with realistic models - results were shitty) Mybe it's because of the dataset, because it's not so much pics in there. If you happens to have a database with clips/pics with that pose - it could help)
@gigsa now it works with realistic models)
@aDDont Thanks man!! It's not perfect and it takes a while to find a good seed but it's the only LORA of this type that exists and I love it!! Give us more!!
效果不是很好，很难出好图