写实人像最难的不是”画得像真人”,而是不让模型陷入”千篇一律网红脸”。这份合集里的 10 套模板,每一套都在风格、光线、镜头三层做了差异化。直接复制,只换主体描述,就能稳定产出不同氛围的高质量写实人像。文末还给出截至 2026 年 6 月各主流模型该跑哪种 Prompt 的建议。
一句话先看
- 写实人像里权重最高的三个词是:镜头规格(
85mm f/1.4)、胶片型号(Kodak Portra 400)、光线方向(north-facing window light)。它们的分量大过任何 “beautiful / stunning” 形容词。 - 截至 2026 年 6 月,纯写实选 Flux 1.1 Pro(毛孔、布料质感最真实);要 4K 极致细节选 Google Nano Banana 2;要灯光与构图的高级感选 Midjourney V7 / V8.1。
- 想让同一张脸贯穿一整组图,用 Midjourney 的 Omni Reference(
--oref+--ow),不要再用旧的--cref——V7 已经不认这个参数了。
写实人像 Prompt 的关键六层
- 人物特征:年龄段、肤色、脸型、五官特征、发型、表情
- 服装与配饰:剪裁、面料、颜色、细节
- 风格与拍摄类型:editorial、fashion photography、cinematic still、documentary 等
- 光线与色温:自然光 / 影棚光 / 黄昏 / 冷调 / 暖调
- 镜头与构图:85mm / 50mm / 35mm,半身 / 全身 / 特写
- 画质与胶片感:Kodak Portra 400、Fujifilm 800Z、digital crisp 等
经验:写实人像里,镜头规格 + 胶片型号 + 灯光方向这三个词的权重,比任何形容词都大。
同一段 Prompt,该跑哪个模型(2026 年 6 月)
同样的描述,在不同工具里出来的效果差别很大。先按需求挑模型,再去抠措辞:
| 模型(当前版本) | 最擅长 | 写实肤质 | 备注 |
|---|---|---|---|
| Flux 1.1 Pro | 真人级写实、“看着像真照片”的需求 | 毛孔、皮肤通透感最强 | 纪实 / 证件照级写实的首选 |
| Google Nano Banana 2 | 极致微观细节、4K 输出 | 布料、皮肤颗粒最锐利 | 要印刷尺寸的质感时用它 |
| Midjourney V7 / V8.1 | 灯光氛围、时尚大片构图 | 很好,略带风格化 | V8.1(2026-04-30 上线)渲染比 V7 快约 4-5 倍 |
| Seedream(即梦) | 风格化人像、画面内中英文字 | 不错 | Prompt 里混排中英文时表现强 |
下面的模板用自然语言加 Midjourney 风格的 --ar 参数写成。在 Midjourney 里可直接用;去掉 --ar 后缀,在 Flux、Nano Banana 2、即梦里同样适用。
10 个可直接复制的 Prompt 模板
1. 高级时尚大片
A 27-year-old East Asian woman with sharp jawline, sleek shoulder-length black hair pulled behind one ear, neutral confident expression, wearing a structured oversized cream blazer over a black silk slip dress
Style: high-end editorial fashion photography, Vogue magazine cover style
Lighting: large softbox front-left, single hair light from above, clean grey backdrop
Lens: 85mm portrait, f/2.0, shot on Hasselblad
Mood: minimalist, confident, slightly cold
Aspect: --ar 4:5
适合:品牌主视觉、个人写真、自媒体头图。把 cream blazer 换成 tailored black trench coat 切换季节感。
2. 自然光肖像(窗边人像)
A 24-year-old woman with soft features, light makeup, wavy long brown hair, gentle smile, wearing a cream cashmere sweater
Style: natural light portrait photography, film aesthetic
Lighting: soft north-facing window light from camera-left, no harsh shadows
Lens: 50mm, f/1.8, shot on Kodak Portra 400
Composition: half body, slight three-quarter angle, simple textured wall behind
Mood: warm, intimate, slow afternoon
Aspect: --ar 4:5
这套是”无脑出片”的基础款。Kodak Portra 400 + north-facing window 几乎不会错。
3. 东方美学人像
A 26-year-old Chinese woman with porcelain skin, almond eyes, refined features, long straight black hair, wearing a modern minimalist hanfu in muted ivory and indigo
Style: contemporary Chinese fashion photography, oriental aesthetic, refined and quiet
Lighting: soft overhead diffused light, gentle backlight, ink-wash backdrop
Lens: 85mm, f/2.2, shot on medium format film
Composition: half body, calm pose, hands gently crossed
Mood: serene, dignified, contemporary oriental
Aspect: --ar 4:5
调参建议:把 modern minimalist hanfu 换成 qipao with subtle floral embroidery,氛围立刻从”国风冷淡风”切换到”老上海”。
4. 电影剧照感
A 30-year-old woman with tired eyes, slightly disheveled long auburn hair, no makeup, wearing a worn navy trench coat over a grey turtleneck
Style: cinematic film still, anamorphic look, In the Mood for Love color grade
Lighting: dim practical streetlight from behind, neon red sign reflected on face
Lens: 35mm anamorphic, slight lens flare, shallow depth of field
Composition: medium shot, looking off-frame, rain-wet city street
Mood: melancholy, narrative, late night
Aspect: --ar 21:9
电影感的本质是 anamorphic + practical light + 一段具体的色调描述。In the Mood for Love color grade 比 cinematic colors 有用十倍。
5. 影棚清冷高冷风
A 25-year-old Asian woman with sharp cheekbones, sleek black ponytail, no smile, wearing a glossy black leather jacket and silver hoop earrings
Style: high-fashion studio portrait, magazine cover quality
Lighting: clamshell lighting with large beauty dish from above, fill from below, deep grey paper backdrop
Lens: 85mm, f/4, shot on Phase One
Composition: tight upper body, dead center, direct eye contact
Mood: confident, edgy, magazine cover
Aspect: --ar 4:5
clamshell lighting 是高级时尚最常用的灯位,能同时让脸部干净、立体感强。
6. 街头纪实感
A 28-year-old woman with natural skin texture, light freckles, brown wavy hair tied messy, casual smile, wearing a vintage denim jacket and white tee
Style: street documentary photography, candid feel, slight film grain
Lighting: overcast afternoon, soft diffused daylight
Lens: 35mm, f/2.8, shot on Fujifilm 400H
Composition: half body, slightly off-center, blurred city street with cafe sign
Mood: easy, real, unposed
Aspect: --ar 3:2
纪实风的关键是 candid + 一个具体环境(cafe sign、subway entrance、bookstore window)。
7. 夏日海边
A 24-year-old woman with sun-kissed skin, sea-breeze tousled hair, soft natural makeup, wearing a cream linen sundress
Style: summer fashion editorial, sun-drenched, soft film grain
Lighting: late afternoon golden hour, warm backlight, gentle lens flare
Lens: 50mm, f/2.0, shot on Kodak Portra 800
Composition: half body, hands holding sun hat, blurred ocean horizon behind
Mood: nostalgic summer, warm and free
Aspect: --ar 4:5
golden hour + backlight + lens flare 是”夏日味”的标准三件套。
8. 深夜城市人像
A 29-year-old woman with smoky eye makeup, dark wine lipstick, wet-look slicked back black hair, wearing a black satin slip dress
Style: night street photography, moody fashion editorial
Lighting: mixed neon (cyan, magenta) reflections from wet pavement, single rim light from above
Lens: 85mm, f/1.4, shot on digital with film emulation
Composition: half body, slightly low angle, blurred neon signs
Mood: dramatic, urban, intoxicating
Aspect: --ar 4:5
这套是 Instagram/小红书最容易”出片”的方向之一。wet pavement 这个词比”夜景”权重大得多。
9. 清新少女文艺风
A 22-year-old woman with bare-face natural look, wispy bangs, light brown shoulder-length hair, gentle close-lipped smile, wearing a white cotton shirt
Style: literary portrait photography, soft natural look
Lighting: overcast soft daylight, no shadow, white wall behind
Lens: 50mm, f/2.0, shot on Kodak Gold 200
Composition: tight portrait, slight tilt of head, hands lightly touching collar
Mood: gentle, fresh, quiet
Aspect: --ar 4:5
bare-face natural look + Kodak Gold 200 几乎是”小清新”的工业标准。
10. 红丝绒复古剧场
A 31-year-old woman with vintage glamour styling, finger waves, deep red matte lipstick, dark winged eyeliner, wearing a structured burgundy velvet dress with pearl earrings
Style: classic Hollywood portrait, 1940s glamour, sepia-rich tones
Lighting: single Fresnel key light front-right, soft shadow on opposite cheek, deep velvet curtain backdrop
Lens: 85mm, f/2.8, shot on medium format film
Composition: tight upper body, three-quarter face, classic pose
Mood: timeless, refined, theatrical
Aspect: --ar 4:5
finger waves + Fresnel key light 是真正的复古关键词。
写实人像的肤质与脸型调参表
| 想要 | 在 Prompt 里加 |
|---|---|
| 真实肤质(不要”塑料感”) | natural skin texture, visible pores, subtle freckles |
| 高级时尚感 | editorial makeup, sculpted highlight, clean matte foundation |
| 减少”AI 脸”味 | imperfect features, asymmetric face, candid expression |
| 东方人像感 | East Asian features, soft eye monolid, refined jawline |
| 西方人像感 | Caucasian features, sharp brow ridge, defined cheekbone |
| 中性高级感 | androgynous features, sharp angular bone structure |
容易踩的坑
- 堆”美感词”:
beautiful, stunning, gorgeous越加越像 AI 脸,反而损失个性。建议删干净,让具体特征说话。 - 没有写镜头:不写
85mm/50mm,模型默认”近距离广角变形”,脸看起来怪。 - 服装含糊:写
pretty dress不如写structured black wool coat with notched lapels。细节越多,模型理解越准。 - 灯光全省略:写实风没灯光就只能赌运气,必加
softbox/golden hour/north window之一。
下一步怎么改得更好
- 加品牌参考:
Vogue editorial style、Phoebe Philo aesthetic、Bottega Veneta campaign mood,模型大多识别得很准。 - 锁人物卡:把”主体描述”块复制到不同 Prompt 里,只换风格 / 场景 / 灯光,做系列。
- 用 Omni Reference 保脸,而不是
--cref:Midjourney 升到 V7 后旧的--cref已经失效(报错或被忽略)。改成把一张清晰正脸图拖进 Omni Reference 框(网页端),或在 Discord 里加--oref [图片链接],再用--ow调强度——默认--ow 100,建议压在400以内,超过容易失控;--ow 70左右能保住脸、又允许换衣服。Flux / SDXL 上则用 15-30 张同一张脸的照片训练 LoRA,锁脸最牢。 - 不要太贪心:一个 Prompt 只描述 1 个人物 + 1 个场景。两个以上人物,建议用 inpaint 后期合成。
FAQ
Q:2026 年哪个模型出的写实人像最像真照片? 要”看着像真照片”,Flux 1.1 Pro 的毛孔和布料质感最真;要 4K 极致细节选 Google Nano Banana 2。Midjourney V7(或更快的 V8.1)灯光和构图最高级,但略带风格化。版本参数见 Midjourney 官方文档。
Q:怎么避免”千篇一律 AI 脸”?
关键不在加形容词,而在三个动作:加具体五官特征(monolid eyes、thin lips、high forehead)、加不完美感(asymmetric face、tired eyes)、删掉所有 “beautiful / perfect / stunning”。
Q:这些 Prompt 能在国内模型(如可灵、即梦、Tencent)上用吗? 基本能。这些国内模型的英文 Prompt 兼容性都不错,但你也可以把每段的结构词翻成中文(“85mm 人像镜头,f/2.0,柯达 Portra 400 胶片质感”),效果差距不大。
Q:怎么让一组人像图风格统一? 锁定 Style、Lighting、Lens 三层,每张只换主体和构图。详见 AI 图片风格不统一怎么办。
Q:人物脸总是有点崩怎么办? 通常是构图 + 镜头问题,不是 Prompt 问题。详见 AI 图片人物脸崩怎么办。
Q:要不要在最后加 8k, masterpiece, best quality?
现代模型(Flux、Midjourney v6+、SDXL)这些词基本无效。把那部分长度让给具体的镜头和光线词更划算。