让AI生成摘要
把“make me clay”这样宽泛的提示词输入基础生成器,扩散去噪器就会默认采用光滑的3D CGI着色。皮肤毛孔完全消失。移动端预设会把自然的面部颗粒感替换成均匀的镜面眩光;你的肖像最终看起来像工厂压模的乙烯基制品。
真实黏土的触觉三要素
真实工作室木偶依靠三个物理标记来骗过微距镜头。跳过它们,材质就会失败。
- 微凹陷: 真正的橡皮泥会保留浅浅的拇指印、指甲划痕,以及脸颊上雕塑刮刀留下的线条。
- 光线消散: 未烘烤的黏土会散射棚拍灯光;其粉质哑光表面会抑制鼻梁处刺眼的白光。
- 骨架接缝: 手工制作的小雕像会显露雕刻的颈部关节、厚重的领口边缘,以及独立珠状眼睛,而不是融合渐变。
打造可信的橡皮泥照片滤镜,需要压制镜面高光,并通过明确的提示词标签注入粗粝的表面颗粒。
准备参考照片以锁定面部相似度
糟糕的输入会很快毁掉身份特征。即便是 Seedream v5.5 或 ChatGPT Image 2.6 这样的精准编辑模型,也需要干净的几何结构来锚定你的面部结构。

锐利黏土轮廓的三条输入规则
- 保持嘴巴闭合: 坚持闭唇。模型追踪起来会容易得多。
- 使用柔和定向光: 使用柔和的侧光。它能为模型提供清晰的深度信息。
- 保持背景简单: 去掉繁杂的背景。简单、有对比的背景能让头发和轮廓保持锐利。
照片就绪快速检查清单
| 输入因素 | 避免 | 推荐 |
| 表情 | 张大嘴笑、面部细节扭曲 | 中性表情或放松的闭唇微笑 |
| 光线 | 强烈的正面闪光、过曝、重度滤镜 | 带可见面部阴影的柔和定向光 |
| 背景 | 繁杂图案、与头部重叠的物体 | 与头发和肩膀清晰分离的简单背景 |
风格1:经典橡皮泥定格动画(Aardman 风格)
在真实的Aardman 工作室布景上,动画师会在每一帧都把真实拇指印按进 Wallace and Gromit 中。完美会扼杀定格动画的魅力。
提示词与实用技巧
我用同一张参考照片测试了 Nano Banana 2.1、GPT Image 2.6 和 Seedream v5.5,看看每个模型如何处理经典橡皮泥外观。

就个人而言,我更喜欢 Seedream v5.5。它产出了最有说服力的手工黏土质感,带有可见的雕塑痕迹、略微不平整的边缘,以及更明显的橡皮泥细节。GPT Image 2.6 的结果比较均衡,而 Nano Banana 2.1 在保留原始面部比例和表情方面做得更好。最佳选择取决于你更在意手工黏土效果还是面部相似度。
以下是我使用的提示词:
plaintext1Transform the input photo into a handcrafted plasticine stop-motion character. 2Preserve the person's recognizable facial features and expression. Recreate the face, hair, and clothing as sculpted colored modeling clay, with matte surfaces, visible fingerprints, hand-shaped details, and slightly uneven edges. 3Rounded sculptural forms, tactile clay texture, miniature stop-motion set photography. 4Avoid glossy plastic, polished CGI, and exaggerated cartoon proportions.
我测试中的几条实用技巧:
- 描述纹理: 明确指定哑光黏土、可见指纹和手塑细节,而不是只依赖“claymation”。
- 平衡相似度与风格: 更强的黏土纹理可能使面部特征变形。把面部比例和表情保留为明确指令。
风格2:光滑聚合物黏土微缩模型(桌面立体模型)
把一张毛茸茸的宠物照片或细节丰富的自拍放进基础移动端预设,毛发就会变成尖刺状的塑料针。柔软橡皮泥在这里显得太凌乱。要复刻烤箱烘烤的 Fimo 或 Sculpey 人偶,需要去除拇指印,并锁定坚实、固化后的边缘。
提示词与实用技巧
这次测试中,我把 Nano Banana 2.1 换成了 Seedream v5.0 Pro。
Seedream v5.5 仍是我的首选。它精准抓住了工艺感:厚实的Q版比例、清晰的刀缝,以及原始的手压工具痕迹。

Seedream v5.0 Pro 精准还原了狗的表情和皮毛,但坚持写实解剖结构,而非Q版比例。GPT Image 2.6 切出了干净边缘,但其流动的毛发读起来像3D渲染,而不是黏土。如果你需要严格的面部相似度,就选 Seedream v5.0 Pro。如果你想要逼真的桌面收藏品,就选 v5.5。
以下是我使用的提示词:
plaintext1Create a handcrafted polymer clay miniature figurine based on the reference photo. 2Preserve the subject's recognizable features, colors, and distinctive details. 3Use smooth, firm, opaque baked polymer clay, with crisp sculpted edges, clean color boundaries, and a subtle satin finish. 4Carefully sculpted details, simplified forms, mounted on a small circular wooden display base. 5Macro product photography, soft studio lighting, neutral tabletop background. 6Avoid rough fingerprints, fuzzy textures, glossy plastic, and polished metallic surfaces.
我测试中的几条实用技巧:
- 锁定固化边缘: 使用
crisp blade seams和smooth cured surfaces,而不是泛泛的黏土标签,以防止柔软橡皮泥般的晕染。 - 平衡相似度与风格:
Miniature chibi proportions会营造玩具美学,但换成1:10 scale realistic figurine可保留面部特征。 - 块面化毛发: 提示
thick sculpted clay slabs,而不是精细毛发细节,以阻止渲染滑向丝滑的3D CGI。
风格3:上釉陶瓷小雕像(窑烧陶器)
在基础AI黏土滤镜中输入“ceramic statue”,生成器就会给你的肖像涂上液态铬或平面乳胶漆。玻璃般的光泽需要深度。
提示词与实用技巧
这次测试我使用同样三个模型:GPT image 2.6、Seedream v5.0 Pro 和 Seedream v5.5。
Seedream v5.0 Pro 既是我的首选,也是最接近真正炻器的匹配。它精准把握了物理感:玻璃般的外套高光、瓷质皮肤,以及锐利、未融化的眼镜。

Seedream v5.5 捕捉到了底座上真实的开片釉裂纹,但厚重的釉层柔化了面部轮廓。GPT image 2.6 紧密跟踪站立轮廓,不过其扁平反射看起来更接近3D树脂,而不是矿物釉。
以下是我使用的提示词:
plaintext1A full-length standing glossy kiln-fired ceramic figurine of a smiling woman wearing thin dark round eyeglasses, an open dark coat, trousers, and boots based on the uploaded reference photo, full-body shot from head to toe, sculpted earthenware pottery, smooth porcelain facial contours, lustrous clear reflective glaze, delicate fine-line crazing pattern on the coat and base, rich mineral pigments, studio rim lighting, boots firmly planted on a round ceramic display pedestal, isolated against a clean neutral studio background
我测试中的几条实用技巧:
- 照片最好无遮挡:使用清晰的站姿,或提示 ceramic bust,以隔离出干净的肩膀。否则,你会得到这样的结果:

- 隔离釉面开片:把开片固定在服装和底座上,防止微裂纹横穿光滑的面部瓷面。
- 固定双脚:明确指定靴子稳稳立在底座上,以防止全身小雕像在膝盖处被裁切。
风格4:暗黑骨架木偶(触感哥特定格动画)
明亮预设会毁掉黑暗奇幻。在像 Guillermo del Toro's Pinocchio 这样的电影的真实摄影棚里,制作者会在带关节的金属骨架上制作木偶,从而在黏土上留下原始机械接缝。
提示词与实用技巧
在这里,Seedream v5.5 是我的最爱。它捕捉到了真实的定格动画物理感:风化的黏土、雕塑感的头发、裸露的黄铜关节,以及琥珀色轮廓光。

Seedream v5.0 Pro 偏向干净的 Laika 风格动画木偶,配有苍白雕塑感面孔和锐利帧感,但缺少粗粝的衰败感。GPT image 2.6 做出了很棒的磨损粗麻布纹理,但保留了写实人类皮肤,读起来像真人奇幻电影,而不是手工木偶。
以下是我使用的提示词:
plaintext1Transform the woman from the uploaded reference portrait into a handcrafted dark fantasy stop-motion puppet, strictly preserving her facial features and thin dark round eyeglasses. 2Convert her skin and clothes into weathered sculptors clay with visible tool marks, exposed brass wire armature seams at the neck and shoulders, and coarse burlap fabrics. 3Set within a physical miniature diorama with moody chiaroscuro lighting, deep shadows, and a sharp amber rim light, 35mm stop-motion film stil
我测试中的几条实用技巧:
- 去除 GPT image 2.6 的皮肤感: 指定
unglazed crumbling clay face和tool marks,防止 GPT 默认生成真人皮肤。 - 隔离黄铜骨架: 将裸露线路限制在
the neck and shoulder joints,这样金属接缝就不会横切眼睛或眼镜。 - 雕刻暗色剪影: 将强烈的明暗对比阴影与
sharp amber rim light搭配,把深色粗麻布服装与微缩背景区分开。 - 区分材质: 提示
frayed burlap stitching以及weathered clay,让织物与雕塑黏土身体部位保持区别。
用定格动画运动模型让静态黏土艺术活起来
真实工作室动画师以每秒12帧“一拍二”拍摄,创造出标志性的阶梯式、手动挪移的顿挫感。
3种场景专属运动提示词配方
将你的静态黏土渲染图作为第1帧上传到视频工作区,并让提示词匹配主体的物理属性:
1. 细微角色生命感(触感聚合物与橡皮泥)
可复制模板: Dramatic 12 fps stop-motion animation, shot on twos. [SCENE DESIGN: Subject & Step Action Chain + Dynamic Light & Shadow + Micro-Material Reaction]. Macro diorama depth, locked static camera, tangible tactile stop-motion weight, zero liquid morphing.
实用替换示例:
plaintext1Dramatic 12 fps stop-motion animation. The puppet reaches stiffly into her frayed burlap pocket, pulls out a small ticking brass pocket watch, and angles it toward her face. A sharp amber streetlamp flickers off-camera, casting stepped, shifting hard shadows across her unglazed clay cheekbones and dark coat. Her sculpted eyes track the dial with rigid mechanical ticks. Macro diorama depth, locked camera, tangible tactile stop-motion weight.
2. 展示转台与光线扫过(上釉陶瓷与瓷器)
可复制模板: Dramatic 12 fps stop-motion animation, shot on twos. [SCENE DESIGN: Rigid Ceramic Subject on Pedestal/Base + Dynamic Lighting Sweep + Stepped Mechanical/Ball-Joint Action & Miniature Prop + Specular Glaze & Crackle Reflection + Zero-Stretch Porcelain Constraint]. Macro diorama framing, locked static tripod camera, intense noir chiaroscuro lighting, authentic hand-manipulated stop-motion stutter, heavy tangible porcelain weight.
实用替换示例:
plaintext1Dramatic 12 fps stop-motion animation, shot on twos, cinematic dark mystery storytelling. A sentient vintage porcelain doll detective, based on the reference image, stands on a round ceramic plinth in a dusty midnight antique cabinet. 2An off-camera flashlight beam sweeps past and fades into darkness. The porcelain figurine suddenly comes to life with crisp, stepped mechanical clicks. She stiffly turns her rigid head to check the shadows, raises a glossy ceramic arm with sharp ball-jointed precision, and reaches inside her crackle-glazed trench coat to draw out a miniature brass magnifying glass. Stepping to the edge of her base with a heavy ceramic tap, she angles the glass downward to inspect an etched cipher on the floor, her spectacles catching a sharp glint of lantern light. Her body remains completely solid fired porcelain—zero rubbery morphing, zero organic flesh stretching. 3Macro diorama framing, locked static tripod camera, intense noir chiaroscuro lighting, authentic hand-manipulated stop-motion stutter, heavy tangible porcelain weight.
3. 经典橡皮泥定格动画(Aardman 风格)
可复制模板: Classic 12 fps plasticine stop-motion animation, shot on twos, British Aardman style. [SCENE DESIGN: Plasticine Subject & Environment + Continuous One-Shot Action Chain + Clay Squash-and-Stretch / Expressive Clay Reactions + Prop & Material Interactions]. Macro diorama framing, locked camera, tactile clay-boil stutter, visible artisan thumbprints, zero CGI morphing.
实用替换示例:
plaintext1Classic 12 fps plasticine stop-motion animation, shot on twos, British Aardman style. A continuous one-shot handheld selfie vlog with zero cuts. Based on the reference image, the handcrafted clay woman holds her smartphone out, excitedly introducing her cozy apartment diorama. She talks with lively stepped mouth shapes, swinging her extended arm smoothly to pan the camera across her miniature living room, pointing enthusiastically toward the wooden dining table and shelf in the background. Her clay curls bounce with authentic stop-motion jitter as she leans into the frame, nodding expressively and flashing a wide, toothy grin to conclude the tour. First-person handheld POV, natural indoor lighting, tactile clay-boil stutter, visible thumbprints, zero CGI morphing.
排查变形与模型失败案例
盲目重掷会浪费额度。当AI黏土滤镜搞砸面部解剖时,你需要有针对性的参数覆盖,而不是从头开始。
将渲染中的视觉伪影与下方根因匹配,并应用相应的参数调整:
| 失败伪影 | 根因 | 直接提示词或参数修复 |
| 融化蜡/蜡烛皮肤 | 过度去噪会剥离物理微纹理。 | 添加:visible thumbprints、matte plasticine (clay) 或 fired porcelain、crackle crazing (ceramic)。 |
| 液态/橡胶般变形 | 视频模型会把镜面光扫过误读为网格弯曲。 | 后缀:tangible physical weight、zero liquid morphing、rigid structure。 |
| 尖桩篱笆牙 | 张嘴笑会模糊微缩牙齿间隙。 | 使用闭嘴照片,或添加:stepped replacement-mouth、Aardman toothy grin。 |
| 360° 网格幻觉 | I2V流程缺少后视参考数据。 | 限制弧度:slow, stepped 15-degree arc。避免360°旋转。 |
| 面部身份漂移 | 风格提示词权重压过源面部几何。 | 降低风格权重;添加:maintain exact eyes, jawline, and spectacles from photo。 |
| 背景糊成一团 | 不受控的背景景深会毁掉微缩尺度。 | 添加:macro diorama framing、tabletop miniature set、shallow DOF。 |
干净的输入永远胜过复杂的提示词工程。在测试材质权重之前,先锁定参考光线和面部几何。一旦你的静态黏土或瓷器渲染在微距裁切下仍保持形状,就直接将其导出到12 fps视频工作流中,以添加物理重量和可触摸的手工感。
从上面的配方中选择一种材质,用你的参考肖像进行单变量测试,并在扩展成完整系列之前优化表面物理效果。






