人物姿势与动作生成的提示词技巧

一句话导读:AI 画人物最怕"站桩式"呆板,这篇讲清如何用提示词驱动自然的姿势和动作。

让 AI 生成一个站着微笑的人很容易,但让它生成"一个正在往咖啡里倒牛奶的人"就难多了。姿势和动作是人物生成中最容易出问题的环节,掌握提示词的写法可以大幅减少返工。

姿势描述的基本原则

AI 模型对姿势的理解来自两个层面:具体的肢体位置描述和整体动作意图。两者都需要写清楚。

反面写法(太模糊):

a woman in a nice pose

正面写法(具体):

a woman leaning against a brick wall, left hand in pocket, right hand holding coffee cup, weight on right leg, head slightly tilted to the left

常用姿势提示词库

站姿

姿势	提示词
侧身倚靠	leaning against [物体], side profile
双手叉腰	hands on hips, confident posture
交叉双臂	arms crossed over chest
单手插兜	one hand in pocket, relaxed stance
回头看	looking back over shoulder, turning around

坐姿

sitting cross-legged on the floor, leaning back on hands

sitting on a bar stool, one leg crossed over the other, elbow resting on counter

动态动作

动态姿势出图难度最高,建议把动作拆成关键帧来描述:

mid-stride, walking forward, left foot stepping, right arm swinging, motion blur on legs

jumping in the air, both feet off ground, arms raised, hair flowing upward, caught in mid-jump

pouring water from a pitcher into a glass, water stream visible, focused expression, hands in action

面部与视线

姿势不只是身体,面部朝向和视线方向同样关键:

• looking directly at camera——正面注视,互动感强
• looking away to the side——侧视,沉思感
• looking down at [物体]——低头,专注感
• gazing into distance——远眺,叙事感
• eyes closed, serene expression——闭眼,宁静感

手部描述

手是姿势中最容易变形的部分。尽量明确写出手在做什么:

both hands holding a book open, fingers visible, natural hand position

right hand reaching for a teacup, fingers slightly curled, delicate grip

动作与道具结合

当姿势涉及与道具的互动时,描述接触方式很重要:

互动	提示词
拿杯子	holding a mug with both hands, fingers wrapped around it
翻书	flipping a page, thumb and index finger pinching paper
打伞	holding an umbrella overhead, tilted slightly to the left
用手机	looking at phone screen, both thumbs typing

减少姿势错误的技巧

1. 避免复杂复合动作——比如"边跑边回头边举杯子",三个动作叠加出错率极高。拆成单一动作更可靠。
2. 用参考姿势词——running pose, yoga warrior pose, ballet arabesque 这类有名称的姿势,模型理解准确度更高。
3. 限制可见肢体数——侧面或半身构图只露出部分肢体,减少手部和关节出错的可能。
4. 分步生成+局部修复——先用简单姿势出全身图,再用配音魔方的图片编辑功能对变形的手或关节做局部重绘。

姿势和动作的描述越具体,画面就越生动。与其反复生成碰运气,不如在提示词里把"谁在做什么、怎么做、手放哪里"说清楚,一次出图的成功率会高得多。