昨天发的工作流更换成另一个 Lora 就可以做照片风格化!
对于服装和姿势的还原也很好。
有点类似米雷经常拍视频的风格,不过没有她手绘的那么精准和传神。
结合视频可以玩起来了,需要调整的提示词我放下面了。
首先需要下载 In Context Lora 中的 portrait-illustration Lora 。
然后通过这个提示词获取图片描述:
为了帮助一位视障艺术家,我们需要详细描述这幅图像的内容,包括从服装、配饰、面部特征、发型、环境、装饰等各个方面。对于出现在图中的人物,虽然不能提及具体姓名,但考虑到艺术家的特殊需求,我们需要以匿名方式描述他们的主要特征(切记不要提及真实姓名)。
通过这个提示词和上面的图片描述获取最终的图像生成提示:
The following is an example prompt: “The image pair showcases the transformation from reality to a stylized interpretation; [IMAGE1] the photo shows a person with a topknot, wearing a cozy yellow sweater and plaid scarf, standing in front of a shop window, while [IMAGE2] the illustrated version highlights the warm tones, adding playful, oversized shapes and bright hues, creating an animated feel with a soft, inviting background.” Please note that this is only a sample prompt and not intended to be the final output. I want you to create a very similar prompt based on the content and style of: "{style:图片描述}", which will go onto the left panel. Based on the following input: "{input1:风格化插图}", craft the right panel accordingly, it needs to say something along the lines of "showing the same content (Clothing, accessories, facial features, hairstyles, environment, decorations etcetc)". NO INTRODUCTION, ONLY THE PROMPT! Go!