AI 动画教程:AI 动画角色一致性提示词包
On this page
当您只需要一张精美的图像时,制作 AI 动画非常简单。
但当该角色需要出现在 10 个、20 个甚至 50 个镜头中时,挑战就来了。
发型变了。衣服突然看起来不一样了。角色的比例在不同场景之间发生变化。有时,同一个角色在不同镜头之间看起来就像一个完全不同的人。
这是 AI 动画中最大的挑战之一:角色一致性。
强大的角色一致性工作流始于提示词。与其在每个镜头中对角色进行不同的描述,不如创建一个可重复使用的角色提示词包,锁定视觉身份,只更改需要更改的元素,例如姿势、表情、镜头角度或环境。
在这篇 AI 动画教程中,我们将向您展示如何构建角色一致性提示词系统,并提供可直接用于角色设计、不同镜头角度、表情、动作、环境和动画镜头的提示词。
为什么角色一致性在 AI 动画中至关重要

在传统动画中,角色三视图(Model Sheet)定义了角色在不同角度和不同情况下的外观。
AI 动画也需要类似的方法。
如果您独立生成每个帧或镜头,模型每次都必须根据您的描述重新创建角色。即使您使用相同的角色名称,模型对角色的理解也可能会有所不同。
例如,这个提示词:
一个年轻的动漫女孩在城市中奔跑。
并没有提供足够的信息来维持一致的角色。
下一次生成可能会给她不同的发型、眼睛颜色、服装或身体比例。
更好的方法是用结构化的描述来定义角色:
一个 19 岁的动漫女孩,留着银色长发,齐刘海,紫罗兰色的眼睛,左眼下方有一颗小泪痣,身材苗条,穿着黑色短款夹克、白衬衫、深色百褶裙、黑色及膝靴,脖子上系着红丝带。
现在您拥有了更强大的视觉身份。
关键在于在您的动画提示词中重复使用相同的角色描述。
角色一致性提示词公式
一个实用的角色一致性提示词可以分为几个层级:
角色身份 + 面部 + 头发 + 身体 + 服装 + 配饰 + 风格 + 姿势 + 表情 + 环境 + 镜头
您不需要为每个镜头重写每个元素。
相反,将您的提示词分为两个部分:
固定角色模块
这在整个项目中基本保持不变。
镜头特定模块
这会因场景而异。
例如:
[角色模块]
19 岁的动漫女孩,留着银色长发,齐刘海,紫罗兰色的眼睛,左眼下方有一颗小泪痣,身材苗条匀称,穿着黑色短款夹克、白衬衫、深色百褶裙、黑色及膝靴,脖子上系着红丝带,左侧戴着独特的银色发夹,干净现代的动漫角色设计。
[镜头模块]
夜晚在下雨的东京街头奔跑,转头回望,眼神坚定,动态动作,低角度镜头,电影级光影。
这种结构使您的提示词更易于管理,并减少了不必要的偏差。
角色一致性提示词包
这是一个可重复使用的提示词包,您可以将其应用于几乎任何 AI 动画项目。
1. 角色基础提示词
首先创建角色的详细参考版本。
Create a full-body character design of [CHARACTER NAME], a [AGE]-year-old [GENDER/CHARACTER TYPE].
Distinctive facial features: [FACE DESCRIPTION].
Hair: [HAIR COLOR, LENGTH, STYLE].
Eyes: [EYE COLOR AND SHAPE].
Skin tone: [SKIN DESCRIPTION].
Body type: [BODY TYPE AND PROPORTIONS].
Outfit: [TOP], [BOTTOM], [SHOES], [OUTERWEAR].
Accessories: [ACCESSORIES].
Keep the character design distinctive, recognizable, and visually consistent.
Clean character silhouette, clear facial features, consistent proportions, polished [ANIME / 3D CARTOON / CINEMATIC / OTHER] style.
Full-body character reference, neutral standing pose, simple background.
示例
Create a full-body character design of Maya, a 19-year-old anime girl.
Distinctive facial features: oval face, small nose, soft jawline, small beauty mark beneath her left eye.
Hair: long silver hair reaching her waist, straight bangs, slightly messy ends.
Eyes: large violet eyes with subtle highlights.
Skin tone: fair warm skin.
Body type: slim athletic build with natural proportions.
Outfit: black cropped jacket, white shirt, dark pleated skirt, black knee-high boots.
Accessories: red ribbon around her neck and a silver hair clip on the left side.
Keep the character design distinctive, recognizable, and visually consistent.
Clean character silhouette, clear facial features, consistent proportions, polished modern anime style.
Full-body character reference, neutral standing pose, simple background.
这成为了您的主角色描述。
保存它。
不要在每次生成新场景时随意更改它。
2. 角色三视图提示词
三视图非常有用,因为它确定了角色从多个方向看过去的外观。
Create a character turnaround sheet for [CHARACTER NAME].
Show the exact same character from:
front view, three-quarter front view, side view, three-quarter back view, and back view.
Maintain identical facial features, hairstyle, hair color, eye color, body proportions, clothing, accessories, and character design in every view.
No redesign, no outfit changes, no hairstyle changes.
Consistent character model, consistent proportions, clean reference sheet, neutral background.
当您的动画包含特写、侧面镜头或角色绕着摄像机移动时,这特别有用。
3. 面部表情提示词包

当表情发生变化时,角色往往会失去一致性。
与其为每种情绪生成一个全新的角色,不如保持身份固定,只修改表情。
开心
Same character, identical facial structure, hairstyle, eye color, outfit, and proportions.
Warm natural smile, bright expressive eyes, relaxed eyebrows, subtle joyful expression.
生气
Same character, identical facial structure, hairstyle, eye color, outfit, and proportions.
Angry expression, narrowed eyes, lowered eyebrows, tense jaw, determined gaze.
悲伤
Same character, identical facial structure, hairstyle, eye color, outfit, and proportions.
Sad expression, slightly lowered eyebrows, watery eyes, subtle trembling lips, emotional but restrained expression.
惊讶
Same character, identical facial structure, hairstyle, eye color, outfit, and proportions.
Surprised expression, widened eyes, raised eyebrows, slightly open mouth, sudden emotional reaction.
恐惧
Same character, identical facial structure, hairstyle, eye color, outfit, and proportions.
Frightened expression, widened eyes, tense facial muscles, slightly open mouth, anxious gaze.
这里重要的短语是:
“same character, identical facial structure, hairstyle, eye color, outfit, and proportions.”(同一角色,相同的面部结构、发型、眼睛颜色、服装和比例。)
它强化了哪些元素应该保持不变。
4. 镜头角度提示词包
角色一致性不仅关乎面部。
从不同的镜头角度观看时,角色也需要保持可识别的比例。
正面视角
Same character, front-facing view, looking directly toward the camera, symmetrical facial features, consistent body proportions.
四分之三侧面视角
Same character, three-quarter view, head turned slightly toward the camera, identical facial structure and hairstyle, consistent proportions.
侧面轮廓
Same character, clean side profile, identical hairstyle, facial proportions, clothing, and accessories.
低角度(仰拍)
Same character, cinematic low-angle shot, camera positioned below eye level, consistent facial structure and body proportions.
高角度(俯拍)
Same character, cinematic high-angle shot, camera positioned above eye level, consistent character proportions and recognizable facial features.
特写
Extreme close-up of the same character's face, identical facial structure, hairstyle, eye color, beauty mark, and distinctive facial features.
5. 姿势提示词包
一旦确立了角色身份,您就可以在不改变角色的情况下改变姿势。
站立
Same character standing naturally, relaxed posture, arms resting naturally at her sides.
行走
Same character walking forward naturally, one foot ahead of the other, relaxed arm movement, natural walking posture.
奔跑
Same character running at full speed, dynamic body posture, arms moving naturally, hair and clothing reacting to motion.
战斗
Same character in a dynamic combat stance, one foot forward, fists raised, focused expression, strong body posture.
坐姿
Same character sitting naturally on a bench, relaxed posture, hands resting naturally, legs positioned comfortably.
回头看
Same character walking away from the camera while turning her head back over her shoulder, surprised expression, natural body rotation.
6. 服装一致性提示词

服装是导致角色走样的另一个常见原因。
AI 模型可能会在不同批次的生成中改变服装的颜色、形状或细节。
使用专门的服装描述:
Maintain the exact same outfit throughout the sequence:
black cropped jacket with silver zipper details,
white fitted shirt,
dark navy pleated skirt,
black knee-high boots,
red ribbon around the neck,
silver hair clip on the left side.
Do not change the clothing colors, materials, design, accessories, or proportions.
对于较长的动画,请描述具体的视觉特征,而不是通用的服装。
对比:
穿着一件黑色夹克
与:
穿着一件短款黑色皮夹克,配有银色斜拉链,领口有两个银色纽扣,左袖上有一个红色小贴片。
第二种描述为模型提供了更多需要保留的信息。
7. 环境提示词包
您的角色应该能够在不同的环境之间移动而不丢失其身份。
城市
Same character standing on a busy futuristic city street at night, neon signs, wet pavement, cinematic atmosphere.
森林
Same character walking through a dense misty forest, tall trees, soft morning light, subtle fog surrounding the character.
学校
Same character walking through a modern Japanese high school hallway, classroom doors, large windows, natural daylight.
屋顶
Same character standing on a city rooftop at sunset, distant skyline, wind moving her hair and clothing naturally.
室内
Same character sitting inside a small modern apartment, warm window light, realistic interior details, cinematic atmosphere.
请注意,环境发生了变化,但角色模块保持不变。
8. 用于 AI 视频的角色 + 动作提示词
当从图像生成过渡到视频生成时,您可以使用 AI 视频生成器 将审核通过的角色视觉效果转化为动画镜头。在您的角色描述中加入动作指令。
一个实用的结构是:
[CHARACTER BLOCK]
[ENVIRONMENT]
[ACTION]
[CAMERA MOVEMENT]
[LIGHTING]
[STYLE]
例如:
Maya, a 19-year-old anime girl with long silver hair, straight bangs, violet eyes, a small beauty mark beneath her left eye, slim athletic build, black cropped jacket, white shirt, dark pleated skirt, black knee-high boots, red ribbon around her neck, silver hair clip on the left side.
She stands alone on a rainy Tokyo street at night.
Maya slowly walks forward through the rain, then suddenly stops and looks over her shoulder. Her silver hair moves naturally in the wind, her jacket reacts subtly to her movement, and raindrops fall continuously around her.
Slow cinematic tracking shot moving backward in front of her, gradually transitioning into a close-up of her face.
Neon reflections on wet pavement, atmospheric night lighting, cinematic anime style, smooth natural motion.
这为视频模型提供了关于身份和动作的信息。
9. 多镜头角色一致性提示词
对于包含多个镜头的 AI 动画,请保持角色模块不变。
镜头 1 — 全景/远景(交代环境)
[CHARACTER BLOCK]
Wide cinematic shot of Maya walking alone through a rainy Tokyo street at night.
Camera slowly tracks backward as she walks forward.
镜头 2 — 中景镜头
[CHARACTER BLOCK]
Medium shot of Maya continuing to walk through the rain.
She looks slightly to the side with a serious expression.
Camera moves alongside her.
镜头 3 — 特写镜头
[CHARACTER BLOCK]
Close-up of Maya's face.
She suddenly notices something in the distance.
Her eyes widen slightly.
Rain falls continuously in the foreground.
Slow push-in toward her face.
镜头 4 — 动作镜头
[CHARACTER BLOCK]
Maya suddenly turns and starts running through the rainy street.
Her silver hair and black jacket move naturally with her motion.
Dynamic tracking camera follows her from behind.
角色描述保持稳定,而镜头描述发生变化。
这是构建一致的 AI 动画最简单的方法之一。
10. 双角色一致性提示词
当多个角色出现在同一个动画中时,请分别定义每个角色。
CHARACTER A:
Maya, 19-year-old anime girl, long silver hair, violet eyes, black cropped jacket, white shirt, dark pleated skirt, red neck ribbon, silver hair clip.
CHARACTER B:
Ren, 21-year-old anime boy, short messy black hair, blue eyes, athletic build, dark green bomber jacket, white T-shirt, black cargo pants, brown boots.
Maintain the exact visual identity of both characters throughout the scene.
Do not swap clothing, hairstyles, facial features, or colors between characters.
然后描述他们的互动:
Maya stands on the left side of the frame while Ren stands on the right.
Maya looks toward Ren with a serious expression.
Ren slowly turns toward Maya.
Maintain clear visual separation between both characters and preserve their exact character designs.
这有助于防止模型将他们的属性混淆在一起。
11. 角色一致性反向提示词(Negative Prompt)
当您的生成系统支持反向提示词时,它会非常有用。
尝试:
different character, different face, different hairstyle, different hair color, different eye color, different outfit, different accessories, altered proportions, inconsistent facial features, distorted face, extra limbs, missing fingers, duplicate character, merged characters, age change, body shape change, clothing change, character redesign
对于动画:
character identity drift, face morphing, changing hairstyle, changing clothes, inconsistent proportions, flickering facial features, unstable eyes, deformed hands, extra limbs, duplicated body parts, inconsistent accessories, sudden character redesign
不过,不要仅仅依赖反向提示词。
强大的正向角色描述和一致的参考图通常更重要。
12. “角色锁定”提示词
对于特别重要的镜头,您可以在提示词末尾添加简短的一致性指令:
Character identity is locked.
Preserve the exact face, hairstyle, eye color, body proportions, outfit, accessories, and distinctive features from the reference.
Only change the pose, expression, camera angle, environment, and action specified in this prompt.
这为生成创建了一个简单的规则:
改变镜头要求的元素。保留其他所有内容。
如何构建角色一致性工作流
提示词包与一致的制作工作流结合使用时效果最佳。
步骤 1:设计角色
从详细的角色提示词开始。
生成几个版本,并选择一个效果良好的角色设计。
不要立即开始生成动画镜头。
首先,确保角色本身是强大且具有辨识度的。
步骤 2:创建角色参考图
创建一张清晰展示角色的参考图像。
理想情况下,包括:
- 全身图
- 正面图
- 四分之三侧面图
- 侧面图
- 面部特写
- 主服装
- 重要配饰
这成为了您的视觉唯一真实水源(Source of Truth)。
步骤 3:创建您的角色模块
将最终的角色设计转化为可重复使用的文本描述。
例如:
[CHARACTER NAME]
[AGE]
[HAIR]
[EYES]
[FACE]
[BODY]
[OUTFIT]
[ACCESSORIES]
[STYLE]
[DISTINCTIVE FEATURES]
保持此模块不变。
步骤 4:创建镜头提示词
现在为每个镜头编写单独的提示词。
与其每次都重新设计您的角色,不如使用:
角色模块
+
镜头描述
+
镜头角度
+
动作
+
光影
这使您的提示词更易于管理。
对于较长的项目,您还可以在 无限画布 上将角色参考、场景描述和镜头创意整理在一起,而不是在不同的文档中分别管理它们。
步骤 5:在动画前生成关键帧
对于复杂的场景,先生成关键视觉图。
检查:
- 面部看起来对吗?
- 服装正确吗?
- 发型一致吗?
- 比例正确吗?
- 配饰都在吗?
- 角色与参考图匹配吗?
如果关键帧有误,请在生成视频之前进行修正。
这可以节省大量的生成时间和算力额度。
步骤 6:让审核通过的视觉效果动起来
一旦角色和构图正确,就将视觉效果转化为动画。
保持动作指令相对集中。
与其要求模型一次性完成所有动作:
她奔跑、跳跃、与三个敌人战斗、转身、拔出剑、看着镜头,然后跳上一栋建筑。
不如从一个更简单的镜头开始:
她向前奔跑,同时摄像机向后追踪。
简单的动作通常更容易控制。
角色一致性提示词模板
这是一个您可以保存并重复使用的通用模板:
[CHARACTER IDENTITY]
Same character: [NAME], [AGE], [CHARACTER TYPE].
Face: [FACIAL FEATURES].
Hair: [HAIR DESCRIPTION].
Eyes: [EYE DESCRIPTION].
Skin: [SKIN DESCRIPTION].
Body: [BODY TYPE].
Outfit: [OUTFIT DESCRIPTION].
Accessories: [ACCESSORIES].
Distinctive features: [UNIQUE FEATURES].
Visual style: [STYLE].
[SCENE]
[ENVIRONMENT DESCRIPTION].
[ACTION]
The character [ACTION DESCRIPTION].
[EXPRESSION]
[EMOTION / FACIAL EXPRESSION].
[CAMERA]
[CAMERA ANGLE + CAMERA MOVEMENT].
[LIGHTING]
[LIGHTING DESCRIPTION].
[CONSISTENCY LOCK]
Preserve the exact character identity, facial features, hairstyle, eye color, body proportions, outfit, accessories, and distinctive features from the reference.
Only change the elements explicitly specified in this shot.
一个完整的 AI 动画示例
让我们把所有内容整合在一起。
假设我们正在制作一个关于 Maya 的简短动漫片段。
角色
Maya is a 19-year-old anime girl with long silver hair reaching her waist, straight bangs, large violet eyes, an oval face, a small beauty mark beneath her left eye, and a slim athletic build.
She wears a black cropped jacket with silver zipper details, a fitted white shirt, a dark navy pleated skirt, black knee-high boots, and a red ribbon around her neck.
She has a distinctive silver hair clip on the left side of her hair.
Modern cinematic anime character design.
镜头 1
Maya walks alone through a rainy Tokyo street at night.
Her long silver hair moves gently in the wind while rain falls around her.
She looks forward with a serious expression.
Wide cinematic tracking shot, camera moving backward in front of her.
Neon signs reflect across the wet pavement.
Atmospheric cinematic anime lighting.
Preserve the exact character identity, face, hairstyle, outfit, proportions, and accessories.
镜头 2
Maya suddenly stops walking.
She slowly turns her head toward something behind her.
Her expression changes from serious to slightly surprised.
Medium three-quarter shot.
Slow camera push-in.
Rain continues falling naturally.
Her hair moves subtly in the wind.
Preserve the exact character identity, face, hairstyle, outfit, proportions, and accessories.
镜头 3
Close-up of Maya's face.
Her violet eyes widen slightly as she realizes someone is following her.
A few strands of silver hair move across her face.
Slow cinematic push-in.
Shallow depth of field.
Rain droplets visible in the foreground.
Dramatic anime lighting.
Preserve the exact facial structure, eye color, hairstyle, beauty mark, and character identity.
镜头 4
Maya suddenly turns around and starts running.
Her silver hair flows dynamically behind her.
Her black jacket and red ribbon react naturally to her movement.
Dynamic tracking shot following her from behind.
Fast but smooth camera movement.
Rain splashes from the pavement as she runs.
Preserve the exact character identity, outfit, hairstyle, proportions, and accessories.
结果会连贯得多,因为每个镜头都是建立在相同的角色基础之上的。
获得更好角色一致性的技巧
1. 不要过度描述角色
描述并非越多越好。
专注于真正定义角色的特征。
头发、眼睛、脸型、服装、配饰和独特标记通常比一大段通用的形容词更有用。
2. 使用具体的标识符
与其使用:
穿着时尚衣服的美丽女孩
不如使用:
银色长发,紫罗兰色眼睛,黑色短款夹克,红色颈部丝带,左侧银色发夹。
具体的视觉细节更容易保留。
3. 保持重要细节的顺序一致
如果您重复生成同一个角色,请保持角色模块的结构一致。
例如:
面部 → 头发 → 眼睛 → 身体 → 服装 → 配饰
不要每次都随机重新排列描述。
4. 避免一次改变太多东西
如果您正在测试一致性,请一次只更改一个变量。
例如:
- 同一角色 + 新姿势
- 同一角色 + 新镜头角度
- 同一角色 + 新环境
- 同一角色 + 新表情
这可以更容易地识别是什么导致了角色走样。
5. 只要有参考图,就使用它
文本提示词描述的是身份,而参考图像提供直接的视觉信息。
如果您的 AI 图像或视频生成器支持角色参考(Character Reference),请在不同镜头中重复使用同一张审核通过的角色参考图。
这对于以下方面特别有用:
- 面部特征
- 头发形状
- 服装细节
- 配饰
- 身体比例
6. 为长项目创建角色设定集(Character Bible)
对于较长的 AI 动画,不要依赖记忆。
创建一个简单的角色设定集,包含:
角色名称:
年龄:
角色定位:
面部:
头发:
眼睛:
身体:
服装:
配饰:
配色方案:
性格:
招牌表情:
招牌姿势:
参考图像:
然后您可以在整个制作过程中重复使用它。
对于更大的项目,AI 智能体 还可以帮助将故事创意、角色、场景和制作步骤整理到更具结构化的工作流中。
角色一致性检查清单
在生成每个新镜头之前,请检查:
- [ ] 相同的角色名称
- [ ] 相同的面部特征
- [ ] 相同的发型
- [ ] 相同的眼睛颜色
- [ ] 相同的身体比例
- [ ] 相同的服装
- [ ] 相同的配饰
- [ ] 相同的独特特征
- [ ] 相同的视觉风格
- [ ] 已附加参考图像(如果可用)
- [ ] 仅更改了预期的镜头元素
- [ ] 镜头和动作描述清晰
如果您的角色一直在变,请回到角色模块,而不是无休止地重新生成镜头。
结语
角色一致性是 AI 生成片段的简单堆砌与连贯的 AI 动画之间最大的区别之一。
我们的目标不是为每个镜头写一个全新的提示词。
而是创建一个稳定的角色身份,然后围绕它构建不同的场景。
从一个详细的角色设计开始。创建一张参考图。将该设计转化为可重复使用的角色提示词。然后将相同的角色模块与不同的表情、姿势、环境、镜头角度和动作相结合。
一旦您拥有了这个系统,您就可以制作从简短的 AI 动漫片段到更长的电影级 AI 电影的所有内容,同时让您的角色在不同场景之间保持极高的辨识度。
您的提示词越有结构,就越容易从单个镜头扩展到整个动画故事。