Back to Details

marketer-localize-campaign

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Localize campaign

营销活动本地化

Take one finished campaign and adapt it to N markets — translated copy, culturally appropriate visuals, right-to-left handling for AR/HE, and text-expansion-safe layouts. Built for consistency (same campaign identity per market) and speed (one batch per locale).
将一个已完成的营销活动适配到N个市场——包括翻译文案、符合当地文化的视觉素材、针对阿拉伯语（AR）/希伯来语（HE）的从右到左（RTL）布局处理，以及适配文案长度变化的安全版式。旨在保证一致性（各市场保持统一的营销活动标识）和高效性（按地区批量处理）。

When to Use

使用场景

  • User has a shipped campaign in one market and needs to launch across 3+ additional locales.
  • "Localize across DE, FR, JP, BR, ES" / "translate + adapt creatives per market" / "RTL version for AR".
  • Master campaign has on-image text, people, or cultural motifs that need swapping per locale.
  • Do NOT use for pure text translation (use an LLM directly) or for a fresh campaign (use 
    marketer-campaign-kit
    first, then this skill).
  • 用户已在一个市场上线营销活动，需要拓展到3个及以上其他地区。
  • 需求示例：“在德国、法国、日本、巴西、西班牙本地化” / “按市场翻译并调整创意素材” / “为阿拉伯语提供RTL版本”。
  • 主营销活动包含图片文案、人物或文化元素，需要按地区替换。
  • 请勿用于纯文案翻译（直接使用LLM即可）或全新营销活动（先使用
    marketer-campaign-kit
    ，再使用本技能）。

Prerequisites

前置条件

Ask up front if the brief doesn't cover it (combine into one message):
  1. Master campaign assets — paths/URLs to the approved creative. What's already finalized (hero, ad variants, email, LP)?
  2. Target locales — language + region (e.g. 
    de-DE
    , 
    fr-FR
    , 
    ja-JP
    , 
    ar-SA
    , 
    pt-BR
    ). Region matters for cultural appropriateness.
  3. Translated copy — is it done? If yes, paste it. If no, note which copywriter/vendor owns it (don't auto-translate ad copy — compliance risk).
  4. Cultural swaps needed — people (face/ethnicity), settings (interior style, streetscape), props (food, gestures)? If the hero has a specific person, does that persist or get replaced per locale?
  5. RTL locales in scope — Arabic, Hebrew, Farsi? Layout needs to mirror, text extends right-to-left.
  6. Deliverable naming — ad-platform uploaders expect 
    {campaign}_{locale}_{size}.webp
    .
Never auto-translate ad copy without a reviewer — legal/compliance owns sign-off.
如果需求说明未涵盖以下内容，请一次性询问：
  1. 主营销活动素材——已获批创意素材的路径/URL。哪些内容已最终确定（主视觉、广告变体、邮件、落地页）？
  2. 目标地区——语言+区域（例如
    de-DE
    、
    fr-FR
    、
    ja-JP
    、
    ar-SA
    、
    pt-BR
    ）。区域对文化适配至关重要。
  3. 翻译好的文案——是否已完成？如果是，请粘贴；如果否，请注明负责翻译的文案人员/供应商（请勿自动翻译广告文案——存在合规风险）。
  4. 需要替换的文化元素——人物（面部/种族）、场景（室内风格、街景）、道具（食物、手势）？如果主视觉有特定人物，是保留还是按地区替换？
  5. 涉及的RTL地区——阿拉伯语、希伯来语、波斯语？布局需要镜像翻转，文案从右到左排列。
  6. 交付物命名规则——广告平台上传工具通常要求格式为
    {campaign}_{locale}_{size}.webp
    。
未经审核者确认，请勿自动翻译广告文案——法律/合规部门负责最终审批。

How to Run

操作步骤

  1. Inventory the master assets. List every creative that needs localizing (hero, 6 ad variants, email, LP hero = 9 per locale × 5 locales = 45 assets).
  2. Collect translated copy per locale BEFORE generating. Don't batch-run with English placeholders — you'll re-render everything.
  3. Choose adaptation strategy per asset:
    • Text-only change (headline on a clean bg): use 
      ideogram-v3
      with the translated copy and same visual.
    • In-place motif swap (person, setting): use an edit model (
      qwen-image-edit-plus
      , 
      flux-kontext-pro
      ) with the original image + a locale-specific prompt.
    • Full regeneration (complex cultural swap): generate fresh from the original prompt + locale guidance, referencing the original as 
      image
      for composition.
  4. Write per-locale manifests. One manifest per locale keeps naming clean and lets you parallelize across locales without collision.
  5. Estimate per locale.
    bash
    gen-ai batch run localize-de.json --dry-run
  6. Run locales in sequence, not in parallel. Review one locale's output end-to-end before kicking off the next — catching a cultural mismatch in DE means fixing the prompt before JP, FR, and BR inherit it.
    bash
    gen-ai batch run localize-de.json -c 4 -o ./out/de-DE
  7. Review with a native speaker per locale. Every locale. No exceptions. Especially for JP (text rendering), AR (RTL + text), and any market with a dedicated country manager.
  8. Package and hand off. Folder per locale, filename includes locale code, 
    results.json
    archived for audit.
  1. 盘点主素材。列出所有需要本地化的创意素材（主视觉、6个广告变体、邮件、落地页主视觉 = 每个地区9个素材 × 5个地区 = 45个素材）。
  2. 生成前收集各地区的翻译文案。不要使用英文占位符批量运行——否则需要重新渲染所有内容。
  3. 为每个素材选择适配策略：
    • 仅修改文案（纯色背景上的标题）：使用
      ideogram-v3
      ，搭配翻译后的文案和原视觉素材。
    • 原位元素替换（人物、场景）：使用编辑模型（
      qwen-image-edit-plus
      、
      flux-kontext-pro
      ），传入原图+地区专属提示词。
    • 完全重新生成（复杂文化替换）：基于原提示词+地区指引重新生成，将原图作为
      image
      参数参考构图。
  4. 编写各地区的清单。每个地区单独一份清单可保持命名规范，且能在各地区并行处理时避免冲突。
  5. 按地区估算成本。
    bash
    gen-ai batch run localize-de.json --dry-run
  6. 按顺序运行各地区任务，而非并行。在启动下一个地区任务前，完整审核前一个地区的输出——如果在德国地区发现文化不匹配问题，可在日本、法国、巴西地区任务启动前修正提示词。
    bash
    gen-ai batch run localize-de.json -c 4 -o ./out/de-DE
  7. 每个地区都由母语使用者审核。所有地区都必须执行，无一例外。尤其针对日本（文案渲染）、阿拉伯语（RTL+文案）以及有专属区域经理的市场。
  8. 打包交付。每个地区单独一个文件夹，文件名包含地区代码，归档
    results.json
    用于审计。

Quick Reference

快速参考

One manifest per locale keeps things tidy. 
{locale}
and 
{name}
placeholders support cross-locale reuse.
json
{
  "defaults": {
    "model": "qwen-image-edit-plus",
    "imageUrls": ["https://cdn-pipeline-output.picsart.com/.../master-hero.webp"],
    "negativePrompt": "low quality, watermark, culturally inappropriate"
  },
  "jobs": [
    { "id": "hero_de-DE",        "prompt": "adapt hero for German market — preserve composition, swap setting to central-European urban, replace overlay text with: \"Jetzt entdecken\"",                     "aspectRatio": "16:9" },
    { "id": "ad_9x16_de-DE",     "prompt": "adapt ad for German market — keep product/subject, swap scene to Berlin street, overlay headline: \"Jetzt entdecken — 20% Rabatt\"",                          "aspectRatio": "9:16" },
    { "id": "ad_1x1_de-DE",      "prompt": "adapt ad for German market — clean background, German headline: \"Jetzt entdecken\", CTA: \"Mehr erfahren\"",                                                  "aspectRatio": "1:1"  },
    { "id": "email_de-DE",       "prompt": "email header for German launch — German copy on-canvas: \"Jetzt entdecken\"",                                                                                   "aspectRatio": "3:1"  },
    { "id": "lp_hero_de-DE",     "prompt": "LP hero for German market — same product, German subtitle: \"Entdecken Sie die neue Kollektion\"",                                                              "aspectRatio": "16:9" }
  ]
}
For RTL locales (AR, HE), mirror layouts explicitly in the prompt: "right-to-left layout, headline reads from right, mirror the hero composition".
For voiceover localization (video campaigns), pair with 
gen-ai generate -m eleven-multilingual-v2
per locale — keep 
voice
ID consistent across locales for brand continuity.
每个地区单独一份清单可保持条理清晰。
{locale}
和
{name}
占位符支持跨地区复用。
json
{
  "defaults": {
    "model": "qwen-image-edit-plus",
    "imageUrls": ["https://cdn-pipeline-output.picsart.com/.../master-hero.webp"],
    "negativePrompt": "low quality, watermark, culturally inappropriate"
  },
  "jobs": [
    { "id": "hero_de-DE",        "prompt": "adapt hero for German market — preserve composition, swap setting to central-European urban, replace overlay text with: \"Jetzt entdecken\"",                     "aspectRatio": "16:9" },
    { "id": "ad_9x16_de-DE",     "prompt": "adapt ad for German market — keep product/subject, swap scene to Berlin street, overlay headline: \"Jetzt entdecken — 20% Rabatt\"",                          "aspectRatio": "9:16" },
    { "id": "ad_1x1_de-DE",      "prompt": "adapt ad for German market — clean background, German headline: \"Jetzt entdecken\", CTA: \"Mehr erfahren\"",                                                  "aspectRatio": "1:1"  },
    { "id": "email_de-DE",       "prompt": "email header for German launch — German copy on-canvas: \"Jetzt entdecken\"",                                                                                   "aspectRatio": "3:1"  },
    { "id": "lp_hero_de-DE",     "prompt": "LP hero for German market — same product, German subtitle: \"Entdecken Sie die neue Kollektion\"",                                                              "aspectRatio": "16:9" }
  ]
}
针对RTL地区（阿拉伯语、希伯来语），需在提示词中明确指定镜像布局：“从右到左布局，标题从右开始阅读，镜像主视觉构图”。
针对视频营销活动的旁白本地化，可搭配每个地区的
gen-ai generate -m eleven-multilingual-v2
——在各地区保持
voice
ID一致，以维护品牌连续性。

Quick Reference

快速参考

Sub-taskModelWhy
In-place text swap on clean bg
ideogram-v3
Only model that reliably renders legible text in multiple scripts (Latin, Cyrillic, JP, AR)
Motif / setting swap, preserve subject
qwen-image-edit-plus
Strong at in-place edits — keeps composition, swaps scene
Instructed edits with reference
flux-kontext-pro
Precise edit-to-prompt adherence, good when you need surgical changes
Full regeneration with cultural guidance
flux-2-pro
Best photoreal; reference the original as 
image
to keep composition
Face/person swap with identity consistency
gemini-3-pro-image
(Nano Banana Pro)		Best at persisting or deliberately swapping a subject across renders
Multilingual voiceover
eleven-multilingual-v2
Consistent voice across locales — set 
voice
ID in 
defaults
Background replacement only
gen-ai change-bg
(subcommand)		Pixel-identical subject, locale-appropriate scene
Confirm IDs with 
gen-ai models
.
子任务模型原因
纯色背景上的原位文案替换
ideogram-v3
唯一能可靠渲染多种文字（拉丁语系、西里尔语、日语、阿拉伯语）清晰文本的模型
元素/场景替换，保留主体
qwen-image-edit-plus
擅长原位编辑——保持构图，替换场景
带参考图的指令式编辑
flux-kontext-pro
严格遵循提示词要求，适合精准修改
带文化指引的完全重新生成
flux-2-pro
最佳写实效果；将原图作为
image
参数参考构图
保持或替换人物身份的人脸/人物替换
gemini-3-pro-image
(Nano Banana Pro)		最擅长在多次渲染中保持或刻意替换主体
多语言旁白
eleven-multilingual-v2
各地区语音一致——在
defaults
中设置
voice
ID
仅替换背景
gen-ai change-bg
（子命令）		主体像素完全一致，场景符合地区文化
通过
gen-ai models
确认模型ID。

Procedure

注意事项

  • Text expansion is real. German runs 30% longer than English, French 20% longer, JP/ZH 30% shorter but character-dense. Leave 25-30% breathing room in text-safe zones before rendering.
  • Never auto-translate ad copy. Compliance owns sign-off. Collect human-reviewed translations, then render.
  • RTL locales mirror the layout. AR/HE need hero composition flipped and headline-reads-right. State it explicitly in the prompt.
  • Culturally appropriate faces/settings. A Berlin scene feels wrong in Tokyo. If the master has a person, decide early: persist (identity consistency via Nano Banana Pro) or swap per market.
  • ideogram-v3
    is the only reliable option for on-image text in non-Latin scripts.     Other models render JP/AR/KR as gibberish. Use it for any text-bearing asset.
  • One locale at a time, reviewed end-to-end. Parallelizing 5 locales hides prompt bugs until you've spent 5x the credits.
  • Native-speaker review is non-negotiable. LLMs miss cultural nuance. Every locale gets eyes on from someone in-market before shipping.
  • Filename locale codes enable direct ad-platform upload. 
    launch_de-DE_9x16.webp
    parses cleanly in Meta / Google / TikTok CSV importers.
  • Keep voice ID consistent across multilingual VO. Same voice across markets = brand continuity. Set 
    voice
    in 
    defaults
    , not per-job.
  • 文案长度变化真实存在。德语比英语长30%，法语长20%，日语/中文短30%但字符密集。渲染前需在文案安全区域预留25-30%的空间。
  • 请勿自动翻译广告文案。合规部门负责最终审批。收集经人工审核的翻译文案后再进行渲染。
  • RTL地区需镜像布局。阿拉伯语/希伯来语用户需要标题从右到左阅读，且主视觉构图翻转。需在提示词中明确说明。
  • 人物/场景需符合当地文化。柏林街景在东京会显得违和。如果主视觉有人物，需提前决定：是保留（通过Nano Banana Pro保持身份一致性）还是按市场替换。
  • ideogram-v3
    是唯一能可靠渲染非拉丁语系图片文案的模型    。其他模型会将日语/阿拉伯语/韩语渲染为乱码。任何包含文案的素材都应使用该模型。
  • 逐个处理地区，完整审核后再进行下一个。并行处理5个地区会隐藏提示词问题，直到消耗了5倍的额度才发现。
  • 母语使用者审核必不可少。LLM会遗漏文化细节。每个地区在上线前都需由当地人员审核。
  • 文件名包含地区代码可支持广告平台直接上传。
    launch_de-DE_9x16.webp
    能在Meta/谷歌/TikTok的CSV导入工具中被正确解析。
  • 多语言旁白需保持语音ID一致。各市场使用相同语音可维护品牌连续性。在
    defaults
    中设置统一的语音ID，而非按任务单独设置。

Pitfalls

常见问题

  • Text overflowing the safe zone in DE / RU / FI. Copy expansion wasn't accounted for in the master layout — leave 25-30% margin before localizing.
  • JP / AR / KR text renders as gibberish. Wrong model. Use 
    ideogram-v3
    for any non-Latin on-image text, not flux/recraft.
  • RTL layout not mirrored. Arabic users get a headline that reads wrong direction + hero composition that feels wrong. Explicit mirror instruction fixes it.
  • Culturally tone-deaf imagery. A Christmas-themed visual in a Muslim-majority market, a left-hand gesture in SEA — run cultural review per market.
  • Translating in the prompt with the LLM. Ad copy goes through compliance; don't have the model "translate this headline to German" mid-prompt. Use human-reviewed translations.
  • Parallelizing all 5 locales upfront. One prompt bug becomes 5 locales of wasted credits. Sequence them; review DE, then kick off FR.
  • Inconsistent voice across locales in VO. Different 
    voice
    IDs per locale breaks brand continuity — set one voice ID across all TTS jobs.
  • Skipping native-speaker review. "The model did fine on German last time" is how you ship a mistranslation.
  • 德语/俄语/芬兰语文案超出安全区域。主版式未考虑文案长度变化——本地化前需预留25-30%的余量。
  • 日语/阿拉伯语/韩语文案渲染为乱码。使用了错误的模型。任何非拉丁语系的图片文案都应使用
    ideogram-v3
    ，而非flux/recraft。
  • RTL布局未镜像翻转。阿拉伯语用户会看到方向错误的标题和违和的主视觉构图。明确的镜像指令可解决此问题。
  • 文化不当的素材。在穆斯林占多数的市场使用圣诞主题素材，在东南亚使用左手手势——每个市场都需进行文化审核。
  • 在提示词中使用LLM翻译。广告文案需经过合规审核；请勿在提示词中让模型“将此标题翻译成德语”。使用经人工审核的翻译文案。
  • 提前并行处理所有5个地区。一个提示词错误会导致5个地区的额度浪费。按顺序处理；审核完德国地区后再启动法国地区任务。
  • 多语言旁白的语音ID不一致。各地区使用不同的
    voice
    ID会破坏品牌连续性——在所有TTS任务中设置统一的语音ID。
  • 跳过母语使用者审核。“上次模型处理德语表现不错”这种想法可能导致错误翻译上线。

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

成本与时间

Rough estimate to localize a 9-asset campaign across 5 locales (45 total renders):
Asset typeCount per localeModelCredits eachPer locale5 locales
Hero + ad + email + LP (motif swap)5
qwen-image-edit-plus
~21050
Text-bearing assets (headlines, CTAs)4
ideogram-v3
~31260
Image subtotal9~22~110 credits
Voiceover (if video campaign)1 per locale
eleven-multilingual-v2
~1-3~2~10
Grand total~120 credits
Wall time (sequential per locale, concurrency 4 within)~3 min~15 min
Full regeneration via 
flux-2-pro
roughly doubles the image credits. Estimate explicitly with 
gen-ai batch run localize-{locale}.json --dry-run
before spending.
将包含9个素材的营销活动本地化到5个地区（共45次渲染）的大致估算：
素材类型每个地区数量模型单次消耗额度单个地区成本5个地区总成本
主视觉+广告+邮件+落地页（元素替换）5
qwen-image-edit-plus
~21050
含文案的素材（标题、CTA）4
ideogram-v3
~31260
图片部分小计9~22~110额度
旁白（若为视频营销活动）每个地区1个
eleven-multilingual-v2
~1-3~2~10
总计~120额度
耗时（按地区顺序处理，单地区并发数4）~3分钟~15分钟
使用
flux-2-pro
完全重新生成素材，图片成本大致翻倍。消耗额度前，请使用
gen-ai batch run localize-{locale}.json --dry-run
进行明确估算。

See also

相关链接

  • gen-ai-use/SKILL.md
    — full CLI reference (flags, model catalog, auth, TTS models)
  • gen-ai-workflows/SKILL.md
    — general multi-step patterns
  • gen-ai-batch/SKILL.md
    — manifest shape, placeholders, concurrency
  • workflows/marketer-campaign-kit/SKILL.md
    — run before this to produce the master campaign
  • workflows/marketer-ad-variant-factory/SKILL.md
    — run winners through localization after A/B tests pick them
  • gen-ai-use/SKILL.md
    — 完整CLI参考（参数、模型目录、身份验证、TTS模型）
  • gen-ai-workflows/SKILL.md
    — 通用多步骤流程模式
  • gen-ai-batch/SKILL.md
    — 清单格式、占位符、并发设置
  • workflows/marketer-campaign-kit/SKILL.md
    — 生成主营销活动需先运行此工具
  • workflows/marketer-ad-variant-factory/SKILL.md
    — A/B测试选出优胜素材后，可通过本工具进行本地化