xingfanxia /jewelry-marketing — 珠宝电商一键营销素材生成器
Source: https://github.com/xingfanxia/AX-skills/tree/main/jewelry-marketing Demo: https://www.bilibili.com/video/BV19hdwBNEDy/（B 站演示视频） License: MIT — fork & adapt freely
💎 完整体验在识川 —— 本 skill 是 AI 电商内容平台「识川 shichuan.ax0x.ai」的珠宝切片。全品类商品分析、图集 / 文案 / 视频生成、目录与批量上架：https://shichuan.ax0x.ai/
One photo in → full Xiaohongshu-ready marketing bundle out. Distilled from the shichuan (识川) production system, narrowed and tuned for jewelry merchants specifically.
Two pipelines, auto-routed
| Input | Pipeline | Output |
|---|---|---|
| 成品珠宝照（项链/戒指/耳环/手镯/吊坠/胸针） | marketing | 12 张 XHS 营销图 + 6 风格文案 + 分析 JSON |
| 原石/散石/裸石照 | design | 8 张设计图（手绘稿 / 3D 渲染 / 佩戴效果 / CAD 技术图...） |
Routing happens automatically from the analysis (input_type=finished_product vs raw_material). User doesn't pick.
Invocation
~/.claude/skills/jewelry-marketing/generate.py PRODUCT_IMAGE.jpg [options]
PEP 723 uv script — no venv setup, uv handles deps on first run.
Options
--output DIR— output directory (default./jewelry_bundle/<timestamp>)--mode auto|marketing|design— override auto-routing (defaultauto)--templates A,B,C— generate only these templates (default = full bundle)--jewelry-type ring|pendant|earring|brooch— design mode subject (default: inferred from analysis)--copy-only— skip image gen, write onlyanalysis.json+copy.md--analyze-only— skip everything, just writeanalysis.json--seller-description TEXT— extra context to inject into analysis (e.g., "天然 GIA 钻 1ct")--concurrency N— parallel image gens (default 4 — bump higher if your OpenAI tier allows)--no-html— skip the Vogue-styleindex.htmllookbook (default: generate + auto-open in browser on macOS)
Marketing template IDs (12)
hero wristNeck gemCutDetail sceneStyleWear flatLayEditorial moodboard wearStyleGrid giftScene priceAnchor meaningCraftPoster knowhowAvoid starTestimonial
Design template IDs (8)
sketch (canonical, others reference it) sketchMulti rendering3d wearing materialBreak moodboard colorDna cad3d
Examples
# Full marketing bundle — finished jewelry product
~/.claude/skills/jewelry-marketing/generate.py ./pearl_necklace.jpg
# → ./jewelry_bundle/<ts>/{analysis.json, copy.md, marketing/01_hero.jpg, ..., marketing/12_starTestimonial.jpg}
# Design suite from a raw stone
~/.claude/skills/jewelry-marketing/generate.py ./loose_stone.jpg --jewelry-type ring
# → ./jewelry_bundle/<ts>/{analysis.json, design/01_sketch.jpg, ..., design/08_cad3d.jpg}
# Just a few templates
~/.claude/skills/jewelry-marketing/generate.py ./ring.jpg --templates hero,wristNeck,priceAnchor
# Copy + analysis only (no image gen — fast/free)
~/.claude/skills/jewelry-marketing/generate.py ./ring.jpg --copy-only
What's in the output bundle
jewelry_bundle/<timestamp>/
├── analysis.json # Product analysis (gemstones, materials, colors, design concept, target audience, scenes, ...)
├── copy.md # 6 XHS copy styles: 闺蜜种草 / 专业测评 / 情绪叙事 / 穿搭攻略 / 文化叙事 / 送礼仪式感
│ # Each style: 5 hooks (titles), 5 selling points, full post (300-500字), 5 tags
├── index.html # Vogue-style lookbook (米白/香槟金/Cormorant Garamond) — auto-opens on macOS; suppress with --no-html
├── marketing/ # finished_product pipeline
│ ├── 01_hero.jpg # 棚拍主图
│ ├── 02_wristNeck.jpg # 上手图/锁骨图（爆款封面）
│ ├── 03_gemCutDetail.jpg # 宝石工艺微距
│ ├── 04_sceneStyleWear.jpg # 场景化佩戴
│ ├── 05_flatLayEditorial.jpg # 单品平铺特写
│ ├── 06_moodboard.jpg # 品牌情绪板
│ ├── 07_wearStyleGrid.jpg # 一饰多搭四宫格
│ ├── 08_giftScene.jpg # 送礼场景海报
│ ├── 09_priceAnchor.jpg # 价格锚点合集
│ ├── 10_meaningCraftPoster.jpg # 寓意工艺大字报
│ ├── 11_knowhowAvoid.jpg # 选购避坑指南
│ └── 12_starTestimonial.jpg # 明星同款/玄学种草
└── design/ # raw_material pipeline (alternative to marketing/)
├── 01_sketch.jpg # 水彩+铅笔手绘稿（canonical, 其他基于此）
├── 02_sketchMulti.jpg # 6视图工程蓝图
├── 03_rendering3d.jpg # 3D 渲染（KeyShot 级）
├── 04_wearing.jpg # 佩戴效果
├── 05_materialBreak.jpg # 材质拆解（成品+裸石+金属）
├── 06_moodboard.jpg # 灵感拼贴
├── 07_colorDna.jpg # 水彩色卡
└── 08_cad3d.jpg # CAD 技术图（5视图+宝石数量表+尺寸标注）
Credentials (only 2 keys needed)
| Var | Purpose |
|---|---|
GEMINI_API_KEY |
Product analysis — Gemini Flash 3, free tier OK |
NEWAPI_API_KEY + NEWAPI_BASE_URL |
GPT Image 2.5 via your NewAPI gateway (URL includes /v1) |
Credentials can sit in env vars or ~/.config/gpt-image/credentials.
Sunburst is the default for finished marketing and design images; Flare is the
HTTP 429 fallback for both generation and edits. --image-variant flare selects
faster drafts and makes Sunburst the fallback. There is no older-model fallback.
Why this skill (vs writing your own prompts)
The 12 marketing prompts are research-backed (千瓜 2024 / 数英 / fxbaogao / 我是产品经理 case studies on 周大福 / HEFANG / 樱桃 / 阑珊珠宝 / 翡翠平安环 / 莫桑钻 / akoya / 古法金 etc.) — each prompt encodes:
- Material-aware lighting (diamond high-contrast, pearl single-soft, jade even-translucency, gold warm-3000K)
- Stone-aware scene routing (古法金→新中式, 钻石→晚宴, 珍珠→咖啡馆白墙)
- Wear-zone detection (项链→颈部锁骨, 戒指→手部, 耳环→侧脸)
- Subcategory-aware copy headlines (翡翠 vs 钻石 vs 珍珠 use different price anchors / gift contexts / avoid-pitfall topics)
- Reference-image dominance handling (moodboards omit ref to avoid product-anchoring)
- Merchant-fillable slots (prices stay as
¥___, no fabricated numbers)
Don't re-derive this. Use the skill.
Reference docs
references/JEWELRY_TEMPLATES.md— what each of the 20 templates looks like, when to use, sample output descriptionreferences/COPY_STYLES.md— the 6 XHS copywriting styles with tone rules and word listsreferences/OUTPUT_BUNDLE.md— bundle directory structure + JSON schema
When NOT to use this skill
- Non-jewelry products (服饰 / 美妆 / 食品) — use generic image-gen + copy tools
- Bulk / batch processing dozens of products at once — wrap this skill in a loop and respect your OpenAI tier's RPM limit
- Brand-locked aesthetics — this skill targets generic XHS-style; if you need Tiffany / Bulgari brand-exact, hand-author prompts