Expandyourfamily(多胎计划）

上传父母的肖像照随机生成孩子的肖像照。 Upload portraits of the parents to randomly generate a portrait of their child.

AI时代，但人才是最大生产力！就是字面意思。 In the AI era, it is people who remain the greatest productive force! This is exactly what it means literally.

• 如题，这个LORA的作用是：上传两张图片代表父母，即面部特征的来源；来生成一个孩子的肖像照，这个肖像的面部特征来源于父母双方的面部特征，如眼睛、耳朵、鼻子、嘴巴、等等，把这些特征放大或缩小，然后某项特征取其一，组合在一起构成孩子的面部特征。 As the title suggests, the function of this LORA is as follows: Upload two images representing the parents, which serve as the sources of facial features; these images are used to generate a child's portrait. The facial features of this portrait are derived from those of both parents, such as the eyes, ears, nose, mouth, and so on. These features are either enlarged or reduced, and for each specific feature, one is selected from either parent. All these selected features are combined to form the child's facial features.

• 生成的孩子肖像，性别随机。（我的训练集包含同等数量的性别结果图，但我实测10几次，发现女孩的比例更高。可能基数不够大，或者玄学） The gender of the generated child's portrait is random. (My training dataset contains an equal number of images with male and female results, but in my over ten actual tests, I found that the proportion of girls is higher. This may be due to an insufficient sample size, or perhaps it's just a matter of chance.)

• 图片要求：推荐示例中的半身肖像正面照，如果人物在图像中占比很低，或者面部朝向很歪，就不会得到好的结果，甚至失败。 Image requirements: It is recommended to use front-facing half-body portraits as shown in the examples. If the person occupies a very small proportion in the image, or the face is tilted significantly, the results will be unsatisfactory, and the generation may even fail.

• 可以不用特别的提示词，只用描述性提示词就可以。类似这样的：“Extract the facial features of both individuals, randomly combine them to generate a new face with a random gender, create a portrait image featuring this new face, and maintain the style of the original images.” You don't need to use special prompt words; just descriptive ones will do. Something like this:“Extract the facial features of both individuals, randomly combine them to generate a new face with a random gender, create a portrait image featuring this new face, and maintain the style of the original images.”

• Lora强度:0.6-1.0 强度越大同质性的可能越大。不绝对。 Lora strength:0.6-1.0 The greater the strength, the higher the possibility of homogeneity. However, this is not absolute.

• 采样器相关:euler simple 20-30steps 引导系数:3.0 Sampler related: Euler simple, 20-30 steps
  Guidance scale: 3.0

• 最后，我的训练集大部分都是来源于亚洲面孔，其他人种的效果未知，后续可能会做。请大胆尝试。 Finally, most of my training dataset is derived from Asian faces, and the effect on other ethnic groups is unknown. This may be addressed in the future. Please feel free to experiment.