使用方法
- 复制安装命令，在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物，并对照来源检查输出。
npx skills add https://github.com/samuraigpt/generative-media-skills --skill muapi-photo-pack-generator开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况，不等于独立质量或安全测试。
按源仓库内容呈现，保留标题、案例、代码、表格、链接以及原文引用的演示图片。
📸 Photo Pack Generator Expert Skill (Identity-Lock Edition)
Transform a single reference photo into a collection of themed images while maintaining extremely high facial identity fidelity.
This skill prioritizes identity preservation first, then applies stylistic transformations like LinkedIn portraits, dating photos, cinematic shots, or fantasy styles.
The system uses Identity Lock Prompting instead of describing the person, preventing the model from generating a new face.
Core Principles
1️⃣ Identity Lock (MOST IMPORTANT)
The generated images must always depict the same person from the reference image.
All prompts MUST include identity lock instructions.
Required identity rules:
- Preserve the exact facial identity from the reference image
- Do not modify eye shape or spacing
- Do not modify nose structure
- Do not modify jawline or chin shape
- Do not modify cheekbones
- Do not modify face proportions
- Identity must remain identical to the reference photo
2️⃣ Vision-First Scene Analysis
The agent MUST analyze the reference image before generation.
However the analysis must NOT describe the person (age, ethnicity, hair etc).
Allowed analysis fields:
- head orientation
- facial angle
- expression
- lighting direction
- framing (portrait / half body / full body)
Example:
Head orientation: slight left tilt Expression: neutral friendly Lighting: soft frontal light Framing: head and shoulders portrait
Agent Execution Flow
Step 1 — Grounding Check
Ensure the user has provided a reference image.
Supported inputs:
- local image
- URL
- uploaded file
Step 2 — Vision Analysis
Extract scene attributes only.
DO NOT describe:
- age
- ethnicity
- beard
- hair
- body type
Identity must come directly from the image.
Step 3 — Category Selection
If the user does not specify a category suggest:
- Tinder
- OldMoney
Step 4 — Prompt Construction
Use the reference image as the identity source.
Preserve the exact facial identity from the reference image.
Identity must remain identical to the reference photo.
Do not change:
- eye shape
- eye spacing
- nose structure
- jawline
- cheekbones
- face proportions
Maintain similar head orientation as the reference.
Scene example:
Outdoor café portrait Soft natural daylight 35mm portrait lens Shallow depth of field Photorealistic skin texture
Step 5 — Negative Prompt
Always include:
different person altered face changed facial features new identity generic face beautified face plastic skin face distortion
Step 6 — Execution
Example:
bash scripts/generate-pack.sh \ --image "./my_face.jpg" \ --category "LinkedIn" \ --identity-lock true \ --num 5
Supported Categories
| Category | Best For | Aesthetic |
|---|---|---|
| Professional | Studio | |
| CEO | Founders | Office |
| Tinder | Dating | Lifestyle |
| OldMoney | Luxury | Estate |
| Cyberpunk | Fantasy | Neon |
| Fitness | Gym | Athletic |
| Travel | Social | Bali/Paris |
| 90s | Retro | Vintage |
| Holiday | Seasonal | Festive |
Guardrails
Fidelity First
Identity preservation is always more important than style.
Never Re-Describe the Person
Avoid prompts like:
"Indian man in his 20s with short hair"
This causes the model to generate a new face.
Identity must come from the reference image only.
Recommended Models
Best results with:
- nano-banana-edit
Result
This system produces:
- consistent identity
- photorealistic images
- multi-style photo packs
- professional outputs
