| name | ai-ugc-factory |
| description | Build enterprise AI ad asset systems and batch AI UGC video production briefs. Use when a user asks for AI UGC ads, product demo videos, ad asset factories, prompt packs, or reusable video production workflows. |
| category | media |
| symbolName | video.badge.waveform |
| seedVersion | 2 |
AI UGC Factory|AI UGC 广告工厂
Use this skill to turn a product or service into a repeatable AI UGC ad production system, not a one-off prompt.
使用这个 skill,把一个产品或服务转成可重复运行的 AI UGC 广告生产系统,而不是一次性 prompt。
Output|输出内容
Produce a compact package:
输出一个轻量生产包:
- product brief|产品 brief;
- asset manifest|素材资产清单;
- reference breakdown|参考视频/参考图拆解;
- 1-3 video concepts|1–3 个视频创意方向;
- beat-level prompt pack|1–2 秒级分镜提示词包;
- generation input rules|生成输入规则;
- QA checklist|质检清单;
- batch-learning plan|批量测试与复盘计划。
Workflow|工作流
1. Anchor product truth|锚定产品事实
Read the product website, source files, screenshots, and user notes. Separate verified facts from assumptions. Do not invent prices, discounts, certifications, bonuses, or performance claims.
读取产品官网、源文件、截图和用户说明。区分已验证事实和推测。不要编造价格、折扣、认证、赠品或效果承诺。
2. Build the asset map|建立素材地图
Rename or label assets by visual function:
按视觉功能重命名或标记素材:
hero-product-image|主产品图
packaging-detail|包装/细节
lifestyle-context|生活场景
feature-proof|功能证明
website-cta|网页/CTA
creator-reference|真人/创作者参考
style-reference|风格参考
contact-asset|联系/转化资产
The manifest should say which assets must be actual model inputs. A path in a prompt is not proof that the model read the image.
素材清单必须说明哪些素材需要作为真实模型输入。只在 prompt 里写路径,不等于模型真的读取了图片。
3. Choose production lanes|选择生产路线
Pick 1-3 lanes that fit the product:
根据产品选择 1–3 条视频生产路线:
- Real-person UGC demo|真人 UGC demo;
- hands/product tabletop demo|手部/桌面产品演示;
- animated explainer|动画解释视频;
- creator reaction / testimonial-style|创作者反应/证言风格;
- founder/product walkthrough|创始人/产品 walkthrough;
- SaaS screen + human narration|SaaS 屏幕录制 + 真人口播。
4. Write beat-level prompts|写分镜级提示词
Every 1-2 seconds should carry a job: hook, product proof, pain point, benefit, objection, credibility cue, visual reset, CTA.
每 1–2 秒画面都应承担一个明确任务:钩子、产品证明、痛点、利益点、异议、可信度、视觉重置或 CTA。
For every segment, state:
每个片段都要写清:
- duration|时长;
- aspect ratio|画幅;
- sound/voice requirements|声音/口播要求;
- exact model input images/videos to attach|实际要附加的模型输入图片/视频;
- visible product requirements|产品可见性要求;
- forbidden claims|禁止声明;
- QA risks|质检风险。
5. Gate before rendering|渲染前确认门槛
Before final video generation, confirm any user-owned style decisions: mascot/character identity, voiceover tone, rhythm, CTA/contact asset, and whether exact captions should be post-composited.
最终生成前,确认所有用户拥有的风格决策:角色/吉祥物、口播语气、节奏、CTA/联系资产,以及精确字幕是否需要后期合成。
6. QA the result|质检成片
Require per segment:
每个片段必须检查:
MODEL_INPUT_IMAGES_ACTUALLY_ATTACHED
PRODUCT_REFERENCE_WAS_IMAGE_INPUT
PROMPT_ONLY_PRODUCT_REFERENCE_USED
PRODUCT_VISIBLE_AND_MATCHES_REAL_ASSET
- duration / resolution / audio present|时长 / 分辨率 / 音频;
- text drift or brand-label issues|文字漂移或品牌标签问题;
- unsupported claims|未证实声明。
If product fidelity or text quality fails in client-facing shots, request a narrow correction pass instead of accepting the render.
如果客户可见镜头中的产品一致性或文字质量失败,应要求窄范围返修,而不是直接接受成片。
Example trigger|触发示例
User: “帮我为这个产品批量生产 AI UGC 广告视频,并沉淀成可复用 workflow。”
Response shape|响应结构:
- Confirm asset sources and style gates|确认素材来源和风格门槛;
- Build manifest|建立素材清单;
- Draft prompt pack|起草提示词包;
- Send rendering work to the media/video owner|把渲染交给媒体/视频 owner;
- QA with proof|带证明做 QA;
- turn stable workflow into a reusable skill only after it succeeds|流程跑通后再沉淀成可复用 skill。