简单来说,提示词的精细程度直接决定Pika输出画面的品质——人群特征、动作能力、空间环境,这三大要素缺一不可,否则AI就容易给出模棱两可或视觉不协调的结果。

精准指定人群特征
在提示词的开头或动作描述之前,直接加上英文限定短语,比如“a 35-year-old East Asian woman with medium build and visible shoulder definition”,这比只写“a woman doing push-ups”要精准得多,能有效控制人物的体型、年龄与人种细节。切记避免使用“fitness model”或“athlete”这类泛称——一旦用了,Pika就会自动套用高饱和肌肉线条加强光打亮的模板化形象,结果很可能跟你想要的普通上班族、产后恢复者或中老年初学者完全不一致。如果还需要突出动作的适配性,可以在后面追加能力描述,比如这样写:“a 60-year-old Caucasian man with slight forward lean, wearing supportive walking shoes, performing seated leg extensions slowly”。
嵌入真实感场景要素
第一个方法:采用“in a [具体空间] + [关键视觉锚点]”的结构来锁定环境。比如“in a sunlit home gym with rubber flooring, a folded yoga mat on the left, and a wall-mounted mirror showing partial reflection”。
第二个方法:在环境描述中加入非静态变量,增强可信度。比如“in a community center fitness room during weekday morning, soft fluorescent lighting, two other blurred participants in background doing light stretching”——注意,务必为背景人物添加blurred或out-of-focus描述,否则Pika极易生成表情僵硬、比例失调的干扰角色。
第三个方法:用材质加光影的组合来替代抽象词汇。比如“in a concrete-floored garage conversion studio with exposed brick walls and diffused north-light through high windows”,这比笼统地说“in a cool gym”有效十倍。
人群与场景联动写法
人群与场景的联动可以按三步走:
第一步,确定核心训练动作,比如standing dumbbell curl;
第二步,绑定执行者身份,比如a Latina woman in her 40s, wearing moisture-wicking capri leggings and holding 8kg dumbbells, slight sweat on temples;
第三步,叠加环境约束,比如inside a compact apartment balcony repurposed as workout zone, potted basil on railing, city skyline softly out of focus behind, natural overcast daylight。
这三步一步都不能少——去掉“apartment balcony”,空间独特性就没了;删掉“slight sweat on temples”,训练感就弱了;省略“city skyline softly out of focus”,背景就会变成AI随机填充的失真贴图。
