不露臉短影音 Pipeline
The Faceless Video Pipeline
YouTube 前 1000 大頻道超過 40% 從不露臉。學會工具鏈與合規紅線——以及為什麼手作 process video 是你最安全的路線。
Over 40% of YouTube's top 1000 channels never show a face. Learn the toolchain, the compliance red lines — and why process video is your safest route.
⚠️ 時效敏感(研究於 2026-07):Sora 消費端已於 2026-04 關閉;YouTube inauthentic content 政策 2025-07-15 生效;TikTok C2PA 自動偵測 2025-01 起。
⚠️ Time-sensitive (researched 2026-07): Sora consumer app closed 2026-04; YouTube's inauthentic content policy live since 2025-07-15; TikTok C2PA auto-detection since 2025-01.
1. 螢幕錄製教學Screen Recording
軟體/AI 工具 demo——螢幕即畫面,零成本。SaaS 首選。
Software/AI demos — the screen is the footage. Zero cost; SaaS favorite.
2. AI 旁白+素材AI Narration + Footage
腳本→AI 配音→stock/AI 畫面。歷史、財經、科普解說最常見。
Script → AI voice → stock/AI visuals. History, finance, science explainer staple.
3. 動畫解說Animated Explainer
The Infographics Show 模式,可建角色 IP。
The Infographics Show model; can build a character IP.
4. Kinetic Typography
文字動態+B-roll+配樂。金句、勵志、語錄類;B2B LinkedIn 影片也適用。
Animated text + B-roll + music. Quotes and motivation; also fits B2B LinkedIn video.
5. ASMR / 過程影片ASMR / Process Video
手作、烹飪、陶藝——拍手與作品即可。手作工藝類最相關,詳見 W12。
Crafts, cooking, pottery — just hands and the work. Most relevant for makers; see W12.
- 統計:YouTube 前 1000 大頻道超過 40% 從不露臉;faceless 內容佔 2025 新創作者變現項目約 38%
- Stats: 40%+ of YouTube's top 1000 channels are faceless; ~38% of new creator monetization in 2025
🎙️ 四家對照與選型邏輯Four Providers & Selection Logic
| 工具 | Tool | 強項 | Strength | 中文表現 / 注意 | Chinese / Caveat |
|---|---|---|---|---|---|
| ElevenLabs | 整體品質最高(v3 停頓/呼吸/語調最自然) | Highest overall quality (v3) | 中文可直接用;輕聲偏軟、捲舌音偏硬;最貴(~$0.30/千字元) | Usable Chinese; priciest (~$0.30/1K chars) | |
| MiniMax TTS | 亞洲語言首選(2025-26 多篇評測) | Top pick for Asian languages | 國語與粵語特別強;300+ 聲音 | Mandarin & Cantonese standout | |
| Azure / Google TTS | 穩定、便宜(~$30/百萬字元)、企業級 | Stable, cheap, enterprise-grade | 有 zh-TW 台灣腔 voices;適合批量 | zh-TW voices available; best for batch | |
| OpenAI TTS | 便宜清晰(~$15/百萬字元) | Cheap and clear | 情緒深度不如 ElevenLabs | Less emotional depth |
- 選型邏輯:品牌主聲音用 ElevenLabs/MiniMax(品質);批量旁白用 Azure/Google(成本差可達 10 倍)
- Selection logic: brand voice = ElevenLabs/MiniMax (quality); batch narration = Azure/Google (up to 10x cheaper)
- 先用各家免費額度、以自己的中文腳本實測再決定
- Test with your own Chinese scripts on free tiers before committing
🎞️ 四種來源Four Sources
- Stock 素材:免費 Pexels/Pixabay;付費 Storyblocks($30/月起)、Artgrid(較不撞片)——勝在真實感(微表情、真實景深)
- Stock: free Pexels/Pixabay; paid Storyblocks, Artgrid — wins on realism (micro-expressions, real depth)
- AI 影片生成 2026 現況:Sora 消費端已關(2026-04-26),API 2026-09-24 終止;現役=Runway Gen-4/4.5(品牌一致性,$95/月級距)、Kling 3.0(人物動作與 lip-sync,最長 180 秒)、Pika($35/月無限生成,性價比)
- AI video in 2026: Sora consumer closed (2026-04-26), API ends 2026-09-24; active = Runway Gen-4/4.5, Kling 3.0, Pika
- Screen recording:OBS 免費——教學類零成本畫面
- Screen recording: OBS is free — zero-cost tutorial footage
- 模板:Canva(快速);After Effects 模板市場(品質高、需技能)
- Templates: Canva (fast); After Effects marketplaces (high quality, needs skill)
- 最佳策略:stock 打底、AI 只補拍不到的鏡頭(超現實、抽象、特定情境)
- Best strategy: stock as the base, AI only for unfilmable shots
🔗 全自動流程與現實The Full-Auto Flow & Reality
Google Sheet
Google Sheet
LLM
LLM
Flux / Kling
Flux / Kling
ElevenLabs
ElevenLabs
Creatomate
Creatomate
必要節點
mandatory
- n8n 官方模板庫有現成 workflow 可直接匯入;可自架省訂閱費、資料自主
- n8n's official template library has ready workflows; self-hosting saves fees and keeps data yours
- 一站式替代:Revid.ai(連結轉影片)、Creatify(UGC 風廣告)、HeyGen(最擬真 avatar、175+ 語言 lip-sync)
- One-stop alternatives: Revid.ai, Creatify, HeyGen (most realistic avatars, 175+ language lip-sync)
- 規模現實:可做到每日 2-3 支,但全自動 ≠ 免審
- Scale reality: 2-3 videos/day is achievable, but full-auto ≠ review-free
📋 YouTube 兩層 + TikTok C2PAYouTube's Two Layers + TikTok C2PA
| 規範 | Rule | 內容 | What it says | 違反後果 | Consequence |
|---|---|---|---|---|---|
| YouTube 揭露義務 | YouTube disclosure | 上傳時用 altered/synthetic 開關揭露「可能誤導以為真實」的 AI 內容 | Flag AI content that could be mistaken for real via the altered/synthetic toggle | 平台強制加不可移除標籤 | Forced, non-removable label |
| YouTube 變現資格(2025-07-15) | YouTube monetization (2025-07-15) | 「inauthentic content」:大量重複、無原創觀點的 AI 影片被拒於 YPP。AI 本身不違規,低原創性才違規 | "Inauthentic content": mass-produced, no-original-perspective AI videos barred from YPP. AI isn't the violation — low originality is | 無法變現 / 被移出 YPP | Demonetized / removed from YPP |
| TikTok(2025-01 起) | 擬真 AI 內容須加 AI label;C2PA Content Credentials 自動偵測,自動標記不可移除 | Realistic AI content needs the AI label; C2PA auto-detection, auto-labels can't be removed | 移除+降觸及+記 strike(2025 下半年移除 5 萬+ 支) | Removal + reach penalty + strike (50K+ videos removed in H2 2025) |
- 重要豁免:AI 寫腳本/caption/hashtag/字幕等「工作流 AI」不需標記——規範對象是畫面與聲音本身
- Key exemption: workflow AI (scripts, captions, hashtags, subtitles) needs no label — the rules target visuals and audio
- 合規三步:真實素材優先 → 擬真 AI 元素主動標 → 每支有原創觀點/加工(旁白觀點、剪輯敘事)
- Three-step compliance: real footage first → proactively label realistic AI → ensure original perspective in every video
🎧 ASMR 手法與結構ASMR Techniques & Structure
- 天生 faceless:拍手、材料、工具、成品——不需 AI 畫面、不碰政策紅線、真實性反而是賣點
- Naturally faceless: hands, materials, tools, the finished piece — no AI visuals, no policy risk, authenticity sells
- 聲音是主角:保留原始環境音(切割/揉捏/刷塗聲)並投資收音;ASMR 觀眾為聲音而來
- Sound is the star: keep the original audio (cutting, kneading, brushing) and invest in recording — ASMR viewers come for the sound
- 視覺手法:慢動作、重複性動作、特寫鏡頭營造 hypnotic 效果
- Visual technique: slow motion, repetitive motion, close-ups for the hypnotic effect
- 結構套 W8 框架:hook 用最戲劇性瞬間(脫模/上釉/成品 reveal)放開頭 1-2 秒 → 倒敘過程 → payoff 成品展示
- Apply the W8 framework: hook with the most dramatic moment (demolding, glazing, the reveal) → rewind through the process → payoff with the finished piece
- Hashtag 生態:#satisfying #asmr #process #handmade + 利基 tag;變現直接——影片即產品展示 → 導流商店
- Hashtags: #satisfying #asmr #process #handmade + niche tags; monetization is direct — the video IS the product demo
| 頻道 | Channel | 模式 | Model | 成績 | Results |
|---|---|---|---|---|---|
| Lofi Girl | 單一動畫角色 IP + 音樂直播——無限重複格式 | One animated character IP + music streams | 1500 萬+ 訂閱,估計月收 $20K-45K | 15M+ subs, est. $20K-45K/mo | |
| The Infographics Show | 純動畫+旁白,歷史到科學 | Pure animation + narration | 1500 萬訂閱 | 15M subs | |
| Cal AI | 純 faceless 解說影片(無創辦人出鏡、無網紅) | Pure faceless explainers (no founder, no influencers) | 34 天做到 $1.5M MRR | $1.5M MRR in 34 days |
- 共同成功因子:格式、語調、主題、品質的極度一致性——觀眾訂閱的是可預期的體驗
- Common factor: extreme consistency in format, tone, topic, and quality
- 搭配 trending sounds 的 faceless 影片分享率高 52%
- Faceless videos with trending sounds see 52% higher share rates
🏦 第一原則:你的自有資產就是內容機器First Principle: Your Own Assets Are the Content Machine
開始比價工具之前,先盤點你已經擁有的畫面來源——多數品牌不需要買 AI 生成,自己的產品/數據/過程本身就是無限素材:
Before pricing tools, inventory the footage you already own — most brands don't need AI generation; your own product, data, or process is infinite material:
- 數位工具/SaaS → 產品操作畫面、生成過程動畫(螢幕錄製,$0)
- Digital tools/SaaS → product screens, generation animations (screen recording, $0)
- 手作/實體 → 製作過程(process video,$0 + 手機)
- Handmade/physical → the making process ($0 + a phone)
- 數據型公司(金融/分析)→ 數據視覺化動畫、圖表生長過程(螢幕錄製儀表板,$0)
- Data companies (finance/analytics) → data-viz animations, growing charts ($0)
- 知識型 → 螢幕上的手寫/白板/簡報動畫
- Knowledge brands → on-screen writing, whiteboards, slide animations
影片結尾 CTA 直接導流自家網站——影片即產品展示,轉換路徑最短。
End-of-video CTAs route straight to your site — the video IS the product demo.
📱 帳號準備(全部免費)Accounts (All Free)
- TikTok Business(免費,送商用音樂庫)、Instagram 專業帳號、YouTube 頻道——短影音鐵三角
- TikTok Business (free, incl. Commercial Music Library), IG professional account, YouTube channel — the video trio
- Pinterest Business(免費,送 Pinterest Trends 關鍵字工具)——視覺/電商/手作類必開
- Pinterest Business (free, incl. Pinterest Trends) — essential for visual/e-commerce/handmade
- X / LinkedIn——B2B、金融、開發者受眾的主場(參考 W3-5 單元的平台選擇公式)
- X / LinkedIn — home turf for B2B, finance, and developer audiences (see the W3-5 unit)
- 平台組合依受眾決定:手作=Pinterest+IG+TikTok;量化/金融=X+YouTube+LinkedIn;SaaS=X+YouTube+PH 生態
- Mix by audience: handmade = Pinterest+IG+TikTok; finance = X+YouTube+LinkedIn; SaaS = X+YouTube+PH ecosystem
🧾 各步驟比價表(2026-07 查證)Step-by-Step Price Table (verified 2026-07)
| 步驟 | Step | $0 免費方案 | $0 Option | 低成本升級 | Paid Upgrade |
|---|---|---|---|---|---|
| 素材 | Footage | OBS(完全免費開源)+手機 | OBS (fully free) + phone | — | |
| 腳本/caption | Scripts | LLM 助手(多數人已有訂閱) | Your LLM assistant | — | |
| 配音 | Voice | Azure TTS F0:每月 50 萬字免費(約 10 小時音訊,zh-TW 腔可用) | Azure TTS F0: 500K chars/mo free | ElevenLabs Starter $6/月(30K credits,含商用權——免費版不含) | ElevenLabs Starter $6/mo (incl. commercial license) |
| 剪輯+字幕 | Editing | Clipchamp(Windows 內建,1080p 官方確認無浮水印)或 DaVinci Resolve(4K 免費);CapCut 免費版手動剪輯無浮水印但自動字幕限約 10 分/專案 | Clipchamp (1080p, no watermark) or DaVinci Resolve (4K free) | CapCut Standard $9.99/月(無限自動字幕) | CapCut Standard $9.99/mo |
| 音樂 | Music | 平台內建 trending sounds(app 內免費)+ Pixabay(商用免費免標注);YouTube Audio Library 只保證 YouTube 內使用 | In-app trending sounds + Pixabay (free commercial) | — | |
| 封面 | Covers | Canva 免費版(做圖無浮水印;只有影片輸出限 720p)或自製 SVG/HTML 模板 | Canva free tier or DIY SVG/HTML templates | — | |
| 排程 | Scheduling | Metricool 免費(1 品牌、50 則/月)或平台原生發布 | Metricool free (50 posts/mo) or native | Buffer $6/頻道/月 | Buffer $6/channel/mo |
🏗️ 四階段演進(不要一步到位)Four Evolution Stages (Don't Skip)
全手動跑通
Manual end-to-end
半自動:腳本+配音
Semi: script+voice
自動合成+人工終審
Auto-render + review
排程發布+數據回饋
Publish + feedback loop
先手動跑 10-20 支確認格式有效(W8-10 的節奏表),再自動化——自動化一條沒驗證的生產線只會更快地量產失敗。
Run 10-20 videos manually first to validate the format — automating an unvalidated pipeline just mass-produces failure faster.
⚙️ n8n 節點分解與工程需求n8n Node Breakdown & Engineering Needs
| 節點 | Node | 技術 | Tech | 工程難度 | Effort |
|---|---|---|---|---|---|
| ① 選題佇列 | ① Topic queue | Google Sheet / Notion DB + n8n trigger(定時或 webhook) | Sheet/Notion + n8n trigger | 低——現成節點 | Low |
| ② 腳本生成 | ② Script gen | LLM API(帶 W8 四段式框架的 system prompt + 品牌語氣範本 + 禁用語清單) | LLM API with the W8 framework prompt + brand voice + banned-word list | 中——prompt 工程是成敗關鍵 | Medium — prompt quality decides everything |
| ③ 配音 | ③ Voice | TTS API(ElevenLabs/Azure);中文需前置斷句與多音字修正步驟 | TTS API; Chinese needs polyphone pre-correction | 低-中——多音字修正表要自建 | Low-Med |
| ④ 畫面 | ④ Visuals | 三選一:(a) 預錄素材庫輪播(最穩)(b) 自製網頁動畫 + headless browser 錄製(最獨特)(c) AI 生成 API(最貴+需標記) | (a) pre-recorded library (b) web animation + headless recording (c) AI gen API | (b) 是差異化所在:中-高 | (b) is the differentiator: Med-High |
| ⑤ 合成渲染 | ⑤ Render | Creatomate/Shotstack API(模板化,~$40/月起)或自架 FFmpeg/Remotion(免費但要寫程式) | Creatomate/Shotstack API or self-hosted FFmpeg/Remotion | API 低/自架中-高 | API low / self-hosted Med-High |
| ⑥ 🧑 人工終審 | ⑥ 🧑 Human review | n8n 審核節點:成品傳 LINE/Slack/Email,按鈕核准才進下一步 | n8n approval node: preview to LINE/Slack, button-gated | 低——必要節點,不可省(W12 政策紅線) | Low — mandatory |
| ⑦ 多平台發布 | ⑦ Publish | YouTube API 直發;TikTok/IG 官方 API 限制多——常見解法是 Metricool/Buffer API 或排程工具中轉 | YouTube API direct; TikTok/IG APIs are restrictive — route via scheduler APIs | 中——平台 API 審核是隱藏成本 | Medium — platform API approval is the hidden cost |
| ⑧ 數據回饋 | ⑧ Feedback | 定時抓各平台成效 → 回寫選題表(哪類 hook 有效)→ 影響②的 prompt | Scheduled metrics pull → back to the topic sheet → informs ② | 中——做到這步才是「系統」 | Medium — this closes the loop |
💸 成本結構與 MVP 建議Cost Structure & MVP Scope
- 固定成本:n8n 自架 $0(VPS 約 $5/月)或 n8n cloud $24/月起;渲染 API ~$40/月起(自架 FFmpeg 則 $0)
- Fixed: n8n self-hosted $0 (VPS ~$5/mo) or cloud from $24/mo; render API ~$40/mo (or $0 self-hosted FFmpeg)
- 變動成本(每支):LLM 腳本 ~$0.01-0.05 + TTS ~$0.1-0.5 + 渲染 ~$0.2-0.5——單支約 $0.3-1,但別忘了 API 成本疊加的教訓(W12 誤區)
- Variable per video: LLM + TTS + render ≈ $0.3-1/video — remember the cost-stacking pitfall
- MVP 建議範圍:Stage 1 半自動(選題表 → LLM 腳本 → TTS 配音自動化,畫面與剪輯仍手動)——工程一個週末可完成,先讓「每支省 30 分鐘」發生
- MVP scope: Stage 1 semi-auto (topics → script → voice), visuals and editing stay manual — a weekend of engineering that saves 30 min/video immediately
- 差異化機會:④(b) 自製網頁動畫+headless 錄製——通用工具做不到你的品牌專屬畫面,這是自動化管線裡最值得投資的自研組件
- The differentiator: ④(b) custom web animations + headless recording — generic tools can't produce your brand-unique visuals
🎓 Q1(W1-12)到此完課。Q2 進入內容系統與自然流量(SEO/AEO、電子報、內容日曆、社群經營)——依 README 規則,展開 Q2 前需先做新鮮度研究。
🎓 Q1 (W1-12) complete. Q2 covers content systems and organic traffic — per the README, a freshness research pass precedes Q2 expansion.