<ul><li>受限于旧时代的照片和影像，我们无法感受到旧时期美女的容颜，但Stable-diffusion让这些成为了可能，例如我前段时间制作的赫本LoRA，近期我打算还原上世纪80到90年代的香港女星们,这个系列将会持续推出，可能会比较慢，但是我尽量会完成它。</li><li>首期为大家带来李嘉欣年轻时的样子，本次训练使用了80张左右的图片，均来源网络，底模型chilloutmix，每张照片训练10步。展示图使用chilloutmix模型，VAE是animevae.pt，由于训练时的图片具有一定年代感，故出图效果可能受到一定影响。</li><li>Limited by the photos and images of the old era, we can't feel the beautiful faces of the old era, but Stable-diffusion makes it possible. For example, Hepburn's LoRA, which I made some time ago, recently plans to restore the Hong Kong female stars from the 1980s to 1990s.This series will continue to roll out, may be slow, but i will try to complete it.</li><li>The first phase brings you the look of lijiaxin when she was young. This training used about 80 pictures, all from the network, the bottom model chilloutmix, each picture training 10 steps. The presentation uses the chilloutmix model, VAE is animevae .pt.Since the pictures during training have a certain sense of age, the effect of plotting may be affected to a certain extent.</li></ul>

80-90' s, hong kong actress program 01：lijiaxin/80-90年代，香港女星计划01：李嘉欣

<ul><li>在本版中，对原版进行培训后，发现部分生成照片存在光线过亮等问题，原版训练图片中删除了一部分，修改了训练的dim值为128，重新进行训练。效果如展示图，感觉照片更接近于现代化，请自行选择使用的版本。我个人更喜欢原版的效果，给人一种复古的感觉。</li><li>In this version, after training the original, it is found that some of the generated photos have the problem of too bright a light, etc. a part of the original training pictures is deleted, and the dim value of training is modified to 128, and the training is performed again. The effect is like a display picture. It feels like the picture is more modern. Please choose your own version. I personally prefer the original effect, giving a sense of retro.</li></ul>

Files with SHA256: f0ba1fbdf286844daffae922f80ddc5d08743d8ff47cd6ef49683c9cb63ae8a0