prosumer-headshot-studio

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Headshot studio

职业头像工作室

Input: one casual phone selfie. Output: a set of professional-grade headshots at platform-correct ratios — LinkedIn polished, ID / passport, editorial / creator portrait, casual founder shot — with the same face across every variant. The bar is "plausibly a real photoshoot" not "obvious AI filter".

输入：一张随意的手机自拍照。输出：一组符合平台比例要求的专业级头像——LinkedIn精修版、证件/护照版、编辑/创作者肖像版、休闲创始人版——所有版本均保持同一张面孔。要求是“看起来像真实拍摄的照片”而非“明显的AI滤镜效果”。

When to Use

使用场景

Solo creator, founder, or job-seeker needs a LinkedIn / speaker bio / About page headshot and doesn't want a photographer
User has one usable selfie (well-lit enough, face clearly visible) and wants 3-5 polished variants
They say "headshot", "LinkedIn photo", "professional profile pic", "polish this selfie", "make this a pro photo", "editorial portrait"
They need the same face across multiple styles (not just one image)

Don't use for: group photos, full-body fashion shoots (different model family), or character consistency across video clips (use

motion-studio

with a generated reference).

独立创作者、创始人或求职者需要用于LinkedIn/演讲者简介/关于页面的头像，但不想找摄影师
用户有一张可用的自拍照（光线充足、面部清晰可见），想要3-5个精修版本
用户提及“职业头像”“LinkedIn照片”“专业个人资料图片”“精修这张自拍照”“把这张变成专业照片”“编辑肖像”
用户需要同一张面孔适配多种风格（而非仅一张图片）

请勿用于： 合影、全身时尚拍摄（不同模特类型）或视频片段中的角色一致性（请使用带生成参考的

motion-studio

）。

Prerequisites

前置要求

Ask before running (combine into one message):

Source selfie — path or URL. Must be frontal, eyes visible, decent lighting. If the source is blurry or dark, recommend re-shooting before generation.
Styles wanted — which variants? Default set is LinkedIn / editorial / casual / ID. Offer mix-and-match.
Wardrobe direction — should the model wear the same outfit as the source, a suggested one ("dark blazer", "neutral knit"), or let the model pick per style?
Aesthetic lean — warm / cool, editorial-magazine / corporate-safe / creative-tech?
Background — neutral studio, office, outdoor, plain color, blurred environment?
Aspect ratios needed — 1:1 for LinkedIn/ID, 4:5 for editorial, 9:16 for reels / stories?

运行前需询问（合并为一条消息）：

源自拍照 —— 文件路径或URL。必须是正面拍摄、眼睛可见、光线良好。如果源照片模糊或过暗，建议重新拍摄后再生成。
所需风格 —— 要哪些版本？默认套装为LinkedIn/编辑/休闲/证件版，可混合搭配选择。
着装方向 —— 模特应穿着与源照片相同的服装，还是建议的款式（如“深色西装外套”“中性针织衫”），或按风格自行选择？
审美倾向 —— 暖色调/冷色调，杂志编辑风/稳妥商务风/创意科技风？
背景 —— 中性工作室、办公室、户外、纯色、模糊环境？
所需宽高比 —— LinkedIn/证件用1:1，编辑用4:5，短视频/故事用9:16？

How to Run

运行步骤

1. INTERVIEW        → confirm source, styles, wardrobe, ratios
2. ENHANCE SOURCE   → upscale the selfie for better identity lock
3. GENERATE         → i2i per style, face reference locked
4. REVIEW           → verify face consistency; check hands, eyes, ears
5. REGENERATE       → one-at-a-time for misses
6. DELIVER          → per-platform folder with correct ratios

Upscale the selfie first. Gemini / Flux i2i models lock identity better from a sharp reference:
bash
```
gen-ai enhance -i selfie.jpg -m topaz-upscale-image --download ./headshots/src
```
Alternatively use
```
picsart-enhance
```
for a lighter pass.

Estimate the batch.

bash

gen-ai batch run headshots.json --dry-run

Generate per style with the enhanced selfie as reference. Face identity locking is the whole game — always pass

-i

bash

gen-ai generate -m gemini-3-pro-image -i ./headshots/src/selfie-hd.png \
  -p "professional LinkedIn headshot of the same person, dark blazer over neutral knit, soft studio lighting, clean charcoal background, shallow depth of field, 50mm lens look, editorial corporate photography" \
  --ar 1:1 --download ./headshots/linkedin

Review each output for identity drift. Common failure modes:
- Face looks subtly different (different nose, jawline, eye shape)
- Skin oversmoothed into uncanny-valley territory
- Hair changed color or style beyond the prompt
- Eyes slightly misaligned
If the face drifted, regenerate with a stronger prompt: "exact same face as the reference image, matching bone structure, skin tone, eye color, and hair exactly".

Deliver in a platform-correct folder layout.

./headshots/
  linkedin/  1:1, 1200×1200
  ig/        4:5, 1080×1350
  story/     9:16, 1080×1920
  id/        1:1, 600×600, neutral background
  editorial/ 4:5, wider crop

1. 沟通确认        → 确认源照片、风格、着装、宽高比
2. 优化源照片   → 放大自拍照以更好地锁定面部特征
3. 生成         → 按风格执行图生图（i2i），锁定面部参考
4. 审核           → 验证面部一致性；检查手部、眼睛、耳朵
5. 重新生成       → 对不合格的版本逐一重新生成
6. 交付           → 按平台分类的文件夹，包含符合比例的文件

先放大自拍照。Gemini / Flux图生图模型从清晰的参考图中锁定面部特征的效果更好：
bash
```
gen-ai enhance -i selfie.jpg -m topaz-upscale-image --download ./headshots/src
```
也可使用
```
picsart-enhance
```
进行轻度优化。

估算批量任务。

bash

gen-ai batch run headshots.json --dry-run

以优化后的自拍照为参考，按风格生成。面部特征锁定是核心——务必传入

-i

参数：

bash

gen-ai generate -m gemini-3-pro-image -i ./headshots/src/selfie-hd.png \
  -p "professional LinkedIn headshot of the same person, dark blazer over neutral knit, soft studio lighting, clean charcoal background, shallow depth of field, 50mm lens look, editorial corporate photography" \
  --ar 1:1 --download ./headshots/linkedin

检查每个输出是否出现面部特征偏移。常见失败情况：
- 面部轻微变形（鼻子、下颌线、眼型不同）
- 皮肤过度磨皮，陷入恐怖谷效应
- 头发颜色或风格与提示词偏差过大
- 眼睛轻微错位
如果面部特征偏移，使用更明确的提示词重新生成：“与参考图完全相同的面孔，匹配骨骼结构、肤色、眼睛颜色和发型”。

按平台要求的文件夹结构交付。

./headshots/
  linkedin/  1:1, 1200×1200
  ig/        4:5, 1080×1350
  story/     9:16, 1080×1920
  id/        1:1, 600×600, neutral background
  editorial/ 4:5, wider crop

Quick Reference

快速参考

json

{
  "defaults": {
    "model": "gemini-3-pro-image",
    "imageUrls": ["./headshots/src/selfie-hd.png"]
  },
  "jobs": [
    {
      "id": "linkedin",
      "prompt": "professional LinkedIn headshot of the same person, dark blazer over neutral top, soft studio lighting, clean charcoal background, shallow DoF, 50mm look, editorial corporate photography, exact same face as reference",
      "aspectRatio": "1:1"
    },
    {
      "id": "id",
      "prompt": "passport-style ID photo of the same person, neutral expression, plain white background, even front lighting, shoulders visible, no shadows, exact same face as reference",
      "aspectRatio": "1:1"
    },
    {
      "id": "editorial",
      "prompt": "editorial portrait of the same person, moody side light, textured dark background, film grain, 85mm lens, shallow DoF, magazine cover composition, exact same face as reference",
      "aspectRatio": "4:5"
    },
    {
      "id": "casual",
      "prompt": "casual founder portrait of the same person, natural window light, cafe background blurred, warm tones, slight smile, 35mm lens, documentary style, exact same face as reference",
      "aspectRatio": "4:5"
    }
  ]
}

json

{
  "defaults": {
    "model": "gemini-3-pro-image",
    "imageUrls": ["./headshots/src/selfie-hd.png"]
  },
  "jobs": [
    {
      "id": "linkedin",
      "prompt": "professional LinkedIn headshot of the same person, dark blazer over neutral top, soft studio lighting, clean charcoal background, shallow DoF, 50mm look, editorial corporate photography, exact same face as reference",
      "aspectRatio": "1:1"
    },
    {
      "id": "id",
      "prompt": "passport-style ID photo of the same person, neutral expression, plain white background, even front lighting, shoulders visible, no shadows, exact same face as reference",
      "aspectRatio": "1:1"
    },
    {
      "id": "editorial",
      "prompt": "editorial portrait of the same person, moody side light, textured dark background, film grain, 85mm lens, shallow DoF, magazine cover composition, exact same face as reference",
      "aspectRatio": "4:5"
    },
    {
      "id": "casual",
      "prompt": "casual founder portrait of the same person, natural window light, cafe background blurred, warm tones, slight smile, 35mm lens, documentary style, exact same face as reference",
      "aspectRatio": "4:5"
    }
  ]
}

Quick Reference

工具参考

Sub-task	Model	Why
Primary — identity-locked headshots	`gemini-3-pro-image` (Nano Banana Pro)	Best face fidelity across restyles, strong prompt adherence
Alternative — creative restyles	`flux-kontext-pro`	Strong i2i edit, faster iteration, slightly less identity-strict
Source enhancement / upscale	`topaz-upscale-image`	Sharpest 2-4× upscale before i2i
Softer enhancement (fewer artifacts)	`picsart-enhance`	Gentle sharpen + color; safer for already-good selfies
Final-pass upscale for print / 4K web	`topaz-upscale-image`	Crisps up the output to retina / print quality
Background swap only (keep face 100%)	`picsart-change-bg`	When the face is perfect but the background is wrong
Background removal (for transparent PNG)	`picsart-remove-bg`	Clean cutout for site / avatar use

Avoid
flux-2-pro
here — it's t2i-dominant and will drift the face more than Gemini or Kontext.

子任务	模型	原因
核心任务——锁定面部特征的头像	`gemini-3-pro-image` (Nano Banana Pro)	在重新造型时面部保真度最高，对提示词的依从性强
备选方案——创意风格重塑	`flux-kontext-pro`	图生图编辑能力强，迭代速度快，面部特征锁定稍弱
源照片优化/放大	`topaz-upscale-image`	图生图前实现最清晰的2-4倍放大
轻度优化（减少伪影）	`picsart-enhance`	温和锐化+调色；适合本身质量较好的自拍照
最终放大（用于印刷/4K网页）	`topaz-upscale-image`	将输出提升至视网膜/印刷级清晰度
仅替换背景（完全保留面部）	`picsart-change-bg`	面部效果完美但背景不合适时使用
移除背景（生成透明PNG）	`picsart-remove-bg`	用于网站/头像的干净抠图

此处避免使用
flux-2-pro
——它以文生图为主，面部特征偏移程度比Gemini或Kontext更大。

Procedure

操作流程

Source quality is everything. A blurry, backlit, or heavily-filtered selfie produces drifted faces. Spend 30 seconds asking the user to re-shoot in daylight before burning credits.
Always upscale the source first. Sharper reference = tighter identity lock downstream.
Use the phrase "the same person" or "exact same face as reference" in every prompt. Models respond to explicit identity cues.
Generate 1-2 variants per style first, verify identity, then scale up. Don't batch 12 before checking the first one.
Keep wardrobe + lighting changes explicit per style — don't leave them ambiguous. "Dark blazer, neutral knit" beats "business clothes".
Check eyes, ears, and hairline — these are where AI headshot drift hides. Ears especially: asymmetric, missing, or fused ears are a giveaway.
Avoid extreme angles or expressions in the prompt — profile shots, wide laughs, tilted-head glamour poses all break identity. Frontal, neutral-to-slight-smile is the safe zone.
Upscale finals to 2048+ for LinkedIn / About-page use. The source was phone quality; the output should not ship at that resolution.

源照片质量是关键。模糊、逆光或过度滤镜的自拍照会导致面部特征偏移。在消耗算力前，花30秒建议用户在日光下重新拍摄。
务必先放大源照片。更清晰的参考图=后续面部特征锁定更精准。
每个提示词中加入“同一人”或“与参考图完全相同的面孔”。AI模型对明确的面部特征提示响应更好。
每种风格先生成1-2个版本，验证面部特征后再批量生成。不要在未检查第一个输出的情况下批量生成12个版本。
明确指定每种风格的着装+光线——不要模糊描述。“深色西装外套，中性针织衫”比“商务服装”效果更好。
检查眼睛、耳朵和发际线——AI头像的特征偏移常出现在这些部位。尤其是耳朵：不对称、缺失或融合的耳朵很容易暴露AI痕迹。
提示词中避免极端角度或表情——侧脸、大笑、歪头的 glamour 姿势都会破坏面部特征一致性。正面、中性至微微笑是安全范围。
最终输出放大至2048像素以上，用于LinkedIn/关于页面。源照片是手机画质，输出不应保持该分辨率。

Pitfalls

常见问题

Face drifts subtly across variants — source too low-res, prompt didn't say "same person", or model wasn't an identity-locking i2i. Fix: upscale source, use Gemini 3 Pro, explicit identity phrasing.
Uncanny plastic skin — over-processed look. Add "natural skin texture, subtle pores, no beauty retouching" to the prompt.
Asymmetric or fused ears — regenerate; don't try to patch in post. Gemini 3 handles this better than older models.
Wrong aspect ratio for LinkedIn — LinkedIn profile is effectively circular cropped from 1:1. Plan compositions with centered face + breathing room on all sides.
Batch-generating before verifying one — if style 1 drifted, styles 2-4 will too. Always verify the first output before fanning out.
Forgetting to upscale the final — shipping a 1024×1024 headshot to LinkedIn looks soft on retina screens. Always final-upscale.

不同版本间面部轻微偏移——源照片分辨率过低、提示词未提及“同一人”或未使用锁定面部特征的图生图模型。解决方法：放大源照片，使用Gemini 3 Pro，加入明确的面部特征提示。
皮肤像塑料般不自然——过度处理的效果。在提示词中加入“自然皮肤纹理，细微毛孔，无美颜磨皮”。
耳朵不对称或融合——重新生成；不要尝试后期修补。Gemini 3比旧模型处理得更好。
LinkedIn宽高比错误——LinkedIn个人资料头像实际是从1:1比例圆形裁剪。构图时要保证面部居中，四周留有足够空间。
未验证单个版本就批量生成——如果第一个风格版本偏移，后续2-4个版本也会偏移。批量生成前务必先验证第一个输出。
忘记放大最终输出——将1024×1024的头像上传至LinkedIn，在视网膜屏幕上会显得模糊。务必进行最终放大。

Verification

验证方法

Run

gen-ai whoami

to confirm authentication, then re-run the failed command with

--debug

运行

gen-ai whoami

确认身份验证，然后添加

--debug

参数重新运行失败的命令。

Cost & time

成本与耗时

Asset	Model	Credits	Time
Source upscale (topaz)	`topaz-upscale-image`	~2	~15s
1× headshot (Gemini 3 Pro, i2i)	`gemini-3-pro-image`	~3	~25s
1× headshot (Flux Kontext, i2i)	`flux-kontext-pro`	~4	~30s
Final upscale per image	`topaz-upscale-image`	~2	~15s
Typical 4-style set (Gemini)	—	~16	~2 min
Typical 4-style set + final upscales	—	~24	~3 min

资产	模型	算力点数	耗时
源照片放大（topaz）	`topaz-upscale-image`	~2	~15秒
1张头像（Gemini 3 Pro，图生图）	`gemini-3-pro-image`	~3	~25秒
1张头像（Flux Kontext，图生图）	`flux-kontext-pro`	~4	~30秒
单张图片最终放大	`topaz-upscale-image`	~2	~15秒
典型4风格套装（Gemini）	—	~16	~2分钟
典型4风格套装+最终放大	—	~24	~3分钟

prosumer-headshot-studio

Original

Translation

Headshot studio

职业头像工作室

When to Use

使用场景

Prerequisites

前置要求

How to Run

运行步骤

Quick Reference

快速参考

Quick Reference

工具参考

Procedure

操作流程

Pitfalls

常见问题

Verification

验证方法

Cost & time

成本与耗时

See also

相关链接