lumenfix
去年 12 月，「AI 圣诞照」刷屏了：把自拍传给 ChatGPT，贴一段「Create an ultra-realistic 8K Christmas portrait」这样的提示词，就能得到一张自己穿着圣诞毛衣站在圣诞树前的照片。网上的提示词合集大多只给文字，我们想知道每条提示词实际会把一张真实的脸变成什么样。
所以我们拿两张照片跑了 8 条提示词：一张单人人像，一张母女合照。下面每张图都是真实生成的结果。我们是通过 API 调用 OpenAI 的 GPT Image 2.5 模型来跑的，不是在 ChatGPT 应用里，所以你在 ChatGPT 里的结果可能不同。每次生成的结果也都会有些差别。
提示词保留英文原文，因为我们测的就是英文版。你也可以用中文写，但效果我们没有测过。
不想写提示词的话，可以直接跳到文末的不用写提示词的办法。
怎么在 ChatGPT 里用这些提示词
- 打开 ChatGPT，附上一张清楚的照片，人脸要对焦清晰、不要太小。
- 从本页复制一条提示词粘贴进去，保留最后一句「保住脸」的话。
- 某张脸或姿势不对，先让它重新生成，不要急着改提示词。
- 下载结果，发出去之前先和原图对比一下。
加在任何提示词后面的「保住脸」句子
下面每条提示词都以这句话结尾，它的作用是让模型尽量保持你本来的样子。你在别处找到的圣诞提示词，也可以加上这句。
Keep the same face, facial features, skin tone, hair and age as in the uploaded photo.两个人或更多人时，用这一版：
Keep both people exactly as they look in the uploaded photo: the same faces, facial features, skin tone, hair and age.我们测试时，加了这句后人脸基本都保住了。姿势不一定：描述动作的提示词（比如装饰圣诞树）会重新安排人的姿势。
单人圣诞人像提示词
四条单人提示词用的都是这张照片。
1. 爆款 8K 圣诞人像
Create an ultra-realistic 8K Christmas portrait of the person in this photo, shot on a 55mm lens with bright, playful holiday lighting. Warm Christmas tree lights and soft bokeh in the background, a cozy knit sweater, natural smile. Keep the same face, facial features, skin tone, hair and age as in the uploaded photo.
脸保住了，姿势变了：她变成坐在沙发上托着下巴。这是去年被复制最多的提示词，出来的效果像客厅里拍的生活照，不像棚拍。
2. 糖果拐杖仙境
Create a glossy, playful Christmas portrait of the person in this photo in a candy cane wonderland: giant red-and-white candy canes, a cream wool coat, soft falling snow and warm string lights, bright studio-style lighting. Keep the same face, facial features, skin tone, hair and age as in the uploaded photo.
糖果拐杖是去年的爆款风格之一。她保留了原来托下巴的姿势，手里多了一根巨大的拐杖糖。
3. 亮片派对人像
Turn this photo into an elegant Christmas studio portrait: a sparkly sequin party outfit, a deep red backdrop with gold bokeh lights, soft glamorous lighting. Keep the same face, facial features, skin tone, hair and age as in the uploaded photo.
这条最接近原图：姿势和构图都没变，只换了衣服和背景。适合做头像。
4. 雪夜街头
Place the person in this photo on a snowy city street at night at Christmas: falling snow, shop windows glowing with warm lights, a long wool coat and a scarf, cinematic 35mm film look. Keep the same face, facial features, skin tone, hair and age as in the uploaded photo.
「cinematic」这个词带来了浅景深、有电影感的背景。脸和姿势都和原图接近。
情侣和全家福圣诞照提示词
四条双人提示词用的都是这张照片。
5. 60 年代复古圣诞贺卡
Turn this photo into a 1960s Christmas card photograph: faded film colors, soft grain, a tinsel Christmas tree and wrapped presents behind them, a white card border, both people in 1960s holiday clothes. Keep both people exactly as they look in the uploaded photo: the same faces, facial features, skin tone, hair and age.
两张脸都和原图接近。金属箔圣诞树、褪色的色调和白边，年代感很足。
6. 壁炉边的亲子睡衣
Place the two people in this photo together on a sofa by a crackling fireplace on Christmas morning, wearing matching red plaid Christmas pajamas and holding mugs of cocoa, warm cozy light. Keep both people exactly as they look in the uploaded photo: the same faces, facial features, skin tone, hair and age.
同款睡衣是经典的全家福创意，杯子和手都画得很干净。
7. 一起装饰圣诞树（抓拍）
Turn this photo into a candid Christmas photo of the two people decorating a Christmas tree together and laughing, warm living-room lights, natural documentary style. Keep both people exactly as they look in the uploaded photo: the same faces, facial features, skin tone, hair and age.
这条对人的改动最大。两人转向圣诞树，脸换了角度。看得出还是同两个人，但要仔细对比。
8. 带「Merry Christmas」的贺卡
Turn this photo into a Christmas card: the two people in festive knit sweaters in front of a snowy log cabin with warm windows, with the words "Merry Christmas" in elegant gold lettering across the top. Keep both people exactly as they look in the uploaded photo: the same faces, facial features, skin tone, hair and age.
我们这次生成的字母拼写正确。但图像模型有时仍会拼错字，打印 50 张之前先读一遍。
能期待什么
- 人脸大多能保住。 加上「保住脸」那句后，两张照片在 8 条提示词里的人脸都认得出来。大合影里的小脸更难，先裁剪照片。
- 姿势会变。 只要提示词描述了动作，人的姿势就会被重新安排。想保留原姿势，就只描述场景和服装，不要写人在做什么。
- 每次结果都不一样。 某次结果不对，先重新生成，再考虑改提示词。
- 文字要检查。 贺卡上的字可能会错，要读一遍。
- 老照片。 有划痕或褪色的老照片，损伤会被带进新图里。先修复，再拿来这里或 ChatGPT 用。
不用写提示词的办法
不想写提示词，可以用 Lumenfix AI 圣诞照片生成器，两个按钮就能搞定最常见的需求。选「圣诞树」或「雪景」，上传一张照片，同样的人就会穿着节日服装出现在场景里。它使用每日免费积分，结果没有水印。风格比上面的提示词少，但什么都不用写。
要打印大尺寸或做贺卡，可以再用图片放大放大。