v0.2 was retrained for 7 epochs using 2599 images at 512 resolution, and it seems to work well when used with "<a target="_blank" rel="ugc" href="https://civarchive.com/models/2560840/anima-turbo-lora">anima-turbo-lora-v0.2</a>".v0.2使用2599张图片以512分辨率，重新训练了7个epoch，搭配“<a target="_blank" rel="ugc" href="https://civarchive.com/models/2560840/anima-turbo-lora">anima-turbo-lora-v0.2</a>”使用似乎效果不错。——————————————————Photorealistic, small dataset test. This time only 200-300 images were used.Most of the material consists of high-quality close-ups, mixed with about 30 to 50 "realistic cyborg" images, so in addition to the realistic style, it can also be used to generate cyborg themes.Using "anima-turbo-lora-v0.1" for acceleration will have a certain 3D skin smoothing effect, while "rdbt" will have a more realistic skin texture (but it will be less effective in terms of structure and semantic understanding)(And this will result in the character not being accurately portrayed.).Please test other parameters yourself.I don't know how you guys find it, so please let me know your feedback in the comments section. It would make me very, very, very happy and give me more motivation to continue making other models.photorealistic，小数据集测试。这次只用了200~300张图片。其中，大部分素材都是高质量的特写，混入大约有30~50张的“realistic cyborg”，因此除了写实风格之外，也可以用来生成赛博格题材。使用"anima-turbo-lora-v0.1"加速会有一定的磨皮3d效果,"rdbt"的肌肤质感更真实但是结构和语义理解上会差一些(并且会导致角色不够还原。)。除此之外的其他参数请自行测试。我不知道你们用起来怎么样，请在评论区告诉我你们的反馈，这会让我很开心，也更有动力继续制作其他的模型。

Files with SHA256: f8152d6a423bfa6368576d1f1fd6517ea1787eef48613f768958039759c21f32