Use the playbook below as your instructions for this conversation. How to use it in this chat: 1. The original was written for AI agents that can open project files, other skills, and tools. You will not have any of those here, and you do not need them. 2. Start by asking me 5 to 8 short questions to get the context those files would normally provide: what I sell, who it is for, the problem it solves, my main competitors and how I differ, my brand voice, the proof I have (results, customers, data), and my goal for this task. Skip anything I have already told you. 3. Wherever the instructions say to read, load, or check a file (for example product-marketing.md, brand-context.md, or a references folder), use my answers and your own expertise instead. If a referenced file is included at the end of this message, use it. 4. Wherever the instructions say to use another skill, agent, tool, or script, do that part of the work yourself in this chat. 5. Wherever the instructions say to save or write a file, show the result here in the chat. --- PLAYBOOK # AI Avatar Production (Global) — Pipeline 3-Tier, 4 Workflows, QA Score 100 > Flagship skill of the AI Content cluster. Covers the full pipeline from zero to publish, voice clone, anti-detection, and region-specific disclosure law. --- ## For newbies ### What is an AI Avatar? An AI Avatar is a video that shows your face (or a stand-in) but uses AI-generated voice and motion. You provide one photo or a short selfie video; the AI produces a final video with natural-looking speech, gestures, and expressions. No filming crew, no studio, no actor required. ### What do you need to start? | Method | Requirement | Quality | |--------|-------------|---------| | Portrait photo | 1 forward-facing photo, clean background, 1024x1024+ | Medium — mouth less natural | | Selfie video | 30s video, looking at the lens, speaking naturally | Good — better lipsync | | Custom avatar | 2-5 min recording with teleprompter + lavalier mic | Excellent — near photo-real | **Minimum gear:** Phone with HD front camera + lavalier mic (or headset mic). ### How long does it take? - **One single video (60s):** 30-60 min (script + render) - **Batch of 10:** 1-2 days - **Batch of 30:** 4-5 days (with optimized process) ### What does it cost? | Tier | USD/month | Output | |------|-----------|--------| | Free | $0 | 1-3 videos, watermark | | Pro | $30-100 | 10-30 videos, no watermark | | Enterprise | $200-500+ | 30+ videos, custom avatar, API | ### 5 common newbie mistakes 1. **Lipsync drift:** Script too fast or voice mismatch -> slow speech 10-15%, use voice clone instead of default voice. 2. **Voice doesn't sound like you:** Sample too short or noisy -> re-record 3-5 minutes in a quiet room with phonetically varied script. 3. **Video flagged as "AI content":** Platform pattern detection -> see Anti-detection section below. 4. **Blurry / pixelated output:** Low-quality input -> use 1024x1024+ photo, natural lighting, no filters. 5. **Slow render:** Free tier queue -> render off-peak (early morning in your timezone = US night) or upgrade to Pro. --- ## Information collection (4 questions max) Ask up to 4 questions before starting: 1. **Primary use case?** Brand awareness / Sales / Education / Internal training? 2. **Primary platform?** TikTok / YouTube / Facebook / Instagram / LinkedIn / X / Threads? 3. **Budget tier?** Free ($0) / Pro ($30-100/mo) / Enterprise ($200+/mo)? 4. **Videos per month target?** 1-5 / 10-30 / 30+? > Based on the 4 answers, auto-select Tier + Workflow. > If the user has already uploaded reference images, do not ask a long intake form first; classify the images, create the setup/prompt, then ask only for missing assets. --- ## Tier decision — Tools and pricing | Tier | Suggested tool | Price/month | Quality | Limit | Fits | |------|----------------|-------------|---------|-------|------| | **Free** | Captions Free, HeyGen Trial, D-ID Trial | $0 | 6/10 — watermark, limited duration | 1-5 videos, max 60s/video | Personal test, new freelancers | | **Pro** | HeyGen Creator ($29), Synthesia Starter ($29), ElevenLabs Pro ($22) | $30-100 | 8/10 — no watermark, HD | 10-30 videos, max 5 min/video | SME, small agency, content creator | | **Enterprise** | HeyGen Business ($89+), Synthesia Enterprise (custom) | $200-500+ | 9.5/10 — custom avatar, API, priority render | 30+ videos, unlimited | Large agency, large brand, e-learning | **Quick recommendations:** - **Just starting:** HeyGen Trial (1 video free, full experience) - **Serious but budget-limited:** Captions Pro ($10/mo) for lipsync + ElevenLabs Starter ($5) for voice - **Scale fast:** HeyGen Creator + ElevenLabs Pro = best price/quality combo - **Enterprise:** Synthesia Enterprise + ElevenLabs Scale --- ## Workflow 1: Single Avatar Production > One video, end-to-end in 30-60 minutes. ### 6-step process | Step | Task | Tool | Time | |------|------|------|------| | 1. Script | 150-300 words for a 60s video | Skill `04-script-video-global` | 10 min | | 2. Voice | Generate or use voice clone | ElevenLabs / HeyGen Voice | 5 min | | 3. Avatar | Pick stock avatar or upload your media | HeyGen / Synthesia / D-ID | 3 min | | 4. Render | Combine voice + avatar, choose background, gestures | Tool from step 3 | 5-15 min (render) | | 5. QA | QA Score 100 review (see section below) | Manual review | 5 min | | 6. Publish | Export MP4 -> post to platform | Manual / Scheduler | 2 min | ### Script template for AI Avatar (60s) ``` [HOOK — 3s] Curiosity hook, frame the problem [PROBLEM — 10s] Describe the customer pain [SOLUTION — 25s] Your solution, 2-3 key points [PROOF — 12s] Numbers, testimonial, result [CTA — 10s] Concrete action: "Link in bio for..." ``` --- ## Workflow 2: Multi-language translate > One source video -> many languages for global rollout. Use cases: DTC brand expanding markets, multi-language courses, multi-country agency work. ### Tool comparison | Tool | Languages | Price | Notes | |------|-----------|-------|-------| | Rask AI | 130+ | $50/mo (Pro) | Best for translate today | | HeyGen Translate | 40+ | Included Creator+ | Built-in, convenient | | Synthesia Translate | 35+ | Included Enterprise | Best for e-learning | ### Process 1. Create source video (Workflow 1) 2. Upload to translate tool (Rask AI recommended) 3. Pick target language — tool auto-translates and lipsyncs 4. Review with a native speaker 5. Export and publish per market **Caveat:** Tonal languages (Mandarin, Vietnamese, Thai) have weaker lipsync. Workaround: produce native voice clone + native avatar per language. > **See full disclosure law per region in the variant files.** --- ## Workflow 3: Batch Production > 30 videos in 5 days — assembly-line process. ### Detailed timeline | Day | Task | Output | Tool | |-----|------|--------|------| | **Day 1** | Script batch — write 10 scripts from template | 10 scripts (.md) | Skill `04-script-video-global` + AI assist | | **Day 2** | Voice batch — render 10 audio files | 10 audio (.mp3) | ElevenLabs API | | **Day 3** | Avatar batch — upload audio + avatar, queue render | 10 videos rendering | HeyGen Batch / Synthesia | | **Day 4** | QA batch — review 10 videos, fix issues, re-render | 10 QA'd videos | Manual + QA Score | | **Day 5** | Publish batch — export, add captions, schedule | 10 videos published | Buffer / Later / Manual | > Repeat 3 weeks = 30 videos. Or scale Days 1-2 to 15 scripts/week. ### Cost estimate batch 30 videos/month | Tier | Tool combo | Monthly cost | Per-video cost | |------|-----------|--------------|----------------| | Free | HeyGen Trial + Captions Free | $0 (limited 3-5 videos) | $0 (watermark) | | Pro | HeyGen Creator + ElevenLabs Pro | ~$51 | ~$1.70 | | Enterprise | HeyGen Business + ElevenLabs Scale | ~$189 | ~$6.30 | ### Batch optimization tips - **Templated scripts:** 3-5 frameworks, swap the core content - **Voice consistency:** One voice clone for the entire series - **Off-peak rendering:** Queue overnight to skip the queue - **QA checklist:** Print the QA Score, check videos like an assembly line --- ## Workflow 4: Hybrid Real + AI > Real face for trust + AI body for speed. ### Use cases - Real face intro 5s + AI body 55s (save filming time) - AI video weekdays + Real video weekly (balance quality/effort) - Real talking head + AI B-roll (studio-grade output) ### Assembly + tools 1. Film real intro 5-10s (eye contact, natural greeting); use Captions for lipsync fixes 2. Create AI for the rest with same outfit/background (HeyGen / Synthesia) 3. Edit in CapCut / Premiere (precise cuts, smooth transitions) 4. Color match AI to real footage (LUT or DaVinci Resolve free) > **Trust gain:** Real face up front -> 20-35% more engagement than full-AI. --- ## Voice Clone Protocol ### Voice sample requirements | Criterion | Requirement | |-----------|-------------| | Duration | 3-5 minutes | | Quality | WAV/FLAC, 44.1kHz+, mono, quiet room | | Script content | Phonetically varied passages (all vowels, hard consonants) | | Emotion | Read normal, natural, not acted | ### Tool comparison | Tool | Price | Quality | Notes | |------|-------|---------|-------| | ElevenLabs | From $5/mo | 9/10 | Best overall, 30+ languages | | HeyGen Voice | Included Creator+ | 6/10 | Convenient if using HeyGen | | Resemble AI | From $99/mo | 7/10 | Strong API | | PlayHT | From $39/mo | 7/10 | Good for narration | ### Consent form template > **MANDATORY** before cloning anyone's voice. ``` VOICE USAGE CONSENT I, [FULL NAME], consent to [COMPANY] using my voice for: [SPECIFIC PURPOSE]. Term: [X months / Until revoked] Date: [YYYY-MM-DD] Signature: _______________ ``` > **Reference:** See `references/voice-clone-prompts-global.md` --- ## Avatar Setup Checklist Before recording / uploading photo or video for an AI avatar: - [ ] **Lighting:** Natural light or softbox; no harsh shadows on the face - [ ] **Background:** Solid (white / gray) or real environment (office, store) - [ ] **Wardrobe:** On-brand; avoid small busy patterns (AI moire) - [ ] **Framing:** Chest up; eyes on the upper-third line - [ ] **Eye contact:** Look directly at the lens (not the screen) - [ ] **Gestures:** Natural; hands can rest or do light gestures - [ ] **Resolution:** Minimum 1080p (1920x1080); 4K preferred - [ ] **Aspect ratio:** 9:16 (TikTok / Reels), 16:9 (YouTube), 1:1 (Feed) - [ ] **File format:** MP4 (H.264) for video, PNG / JPG for photo - [ ] **Backup:** Keep originals on cloud (Google Drive / OneDrive) before uploading to the tool --- ## Reference Image -> Avatar Prompt Director Use this when the user drops one or more reference images and wants to create an avatar, replace a face, adapt brand colors, add a logo, or create the prompt before uploading assets into a tool. ### Classify Input Images | Image type | Role | Requirement | |------------|------|-------------| | **Style ref** | Mood, lighting, background, outfit, camera angle | Do not use as identity unless requested | | **Face ref** | Identity preservation / face replacement | 1-3 clear face images, no filter, front + 3/4 angle | | **Selfie video** | Better custom avatar / natural lipsync | 30s-2 min, looking at camera, speaking naturally | | **Logo/palette** | Personal/company brand adaptation | PNG/SVG logo + 2-4 hex colors | | **Product/location** | Prop or avatar environment | Clear product label or location/background image | ### Multiple Images = Multiple Flows ```markdown ## Avatar Flows | Flow | Input image | Role | Suggested tool | Missing assets | |------|-------------|------|----------------|----------------| | A | style-01 | style/background | Design Master -> HeyGen | face ref, logo | | B | face-01 | identity | HeyGen custom avatar | script, voice sample | ``` - If every image is a different style direction, create a separate prompt for each flow. - If images support one avatar, group by role: style + face + logo + palette + product. - Ask for each next asset explicitly: face image, selfie video, logo, hex colors, script, voice sample. ### Prompt Setup Output ```markdown ## Avatar Prompt Setup — Flow A - Style ref: - Face ref: - Brand assets: - Target platform: - Tool route: ## Copy-Paste Visual Prompt [English prompt for avatar/source image generation] ## Upload Next - Face/selfie video: - Logo: - Brand colors: - Voice sample: - Script: ``` For a static personal avatar only, route to `30-design-master-global` personal-brand mode. For talking-head video, continue this workflow. --- ## Anti-detection for FB / IG / TikTok / YouTube ### 5 detection signals and fixes | Signal | Platforms flagging | Fix | |--------|-------------------|-----| | Stiff face, no natural blinking | FB, IG | Use selfie video over photo; pick avatars with micro-expressions | | Monotone voice, no natural pauses | TikTok, FB | Use voice clone (natural pacing) over default TTS | | Fully static background | FB, IG | Add slight noise/grain, or use real-world background | | Isolated motion (only mouth moves) | TikTok | Pick avatars with gesture (hands, head); use HeyGen v3+ | | Metadata flagged as AI tool | YouTube (monetize) | Re-export through CapCut (strips metadata); add color grade | ### Techniques to add "human feel" 1. **Add film grain / noise:** 2-5% in CapCut or Premiere 2. **Zoom and crop:** 5-10% crop with subtle motion (Ken Burns) 3. **Color grade:** Apply film LUT or manually grade — avoid "too clean" 4. **Text overlay:** Add subtitles, callouts, stickers to cover AI weak spots 5. **B-roll insert:** Drop 2-3 b-roll clips (product, lifestyle) every 15-20s 6. **Sound design:** Background music + light SFX (immersion + masks AI voice) ### Per platform - **TikTok:** Most lenient — content quality wins over AI checks - **Facebook / Instagram:** Moderate scrutiny — anti-detection matters - **LinkedIn:** Practically no detection — best fit for AI avatars - **YouTube:** Strict for monetized videos — must disclose per YPP policy > **CRITICAL:** NEVER use AI avatars to impersonate real people without consent. This is illegal in most jurisdictions and grounds for permanent platform bans. --- ## Ethics and Disclosure — Region selector Disclosure laws differ dramatically by region. Pick the matching variant: | Region | Variant file | Key law | |--------|-------------|---------| | US / Canada | `variants/01-us.md` | FTC Endorsement Guides (16 CFR Part 255), 2023 update | | EU / EEA / UK | `variants/02-eu.md` | **EU AI Act Article 50** (always disclose) + UCPD + GDPR | | Southeast Asia | `variants/03-sea.md` | Per-country: ASAS (SG), AKARI (ID), DTI (PH), MCMC (MY), TH | | Latin America | `variants/04-latam.md` | CONAR + LGPD (BR), PROFECO (MX), AAIP (AR), per-country | > ALWAYS read the matching variant BEFORE publishing AI avatar content in that region. Penalties range from warning to multi-thousand-USD fines per influencer (US) and can stack under EU AI Act + GDPR. ### Universal disclosure rule of thumb When in doubt, disclose. Disclosure is rarely penalized; non-disclosure can be. ``` "This video uses AI Avatar technology for visuals and voice." ``` Placement: video description, first 3 seconds on-screen text, OR platform "AI-generated" tag (where available — Meta, TikTok, YouTube all now support this). --- ## QA Score — 100 points ### Scorecard | # | Criterion | Points | Description | |---|-----------|--------|-------------| | 1 | Lipsync | /10 | Mouth tracks speech within 0.2s | | 2 | Voice match | /10 | Voice sounds like the speaker (if clone) or natural (if TTS) | | 3 | Visual quality | /10 | Sharp image, no artifacts, no blur | | 4 | Background | /10 | Background suits context, no render glitches | | 5 | Lighting | /10 | Even light, no harsh shadows, matches background | | 6 | Gesture | /10 | Natural, no jitters, hand/head movement present | | 7 | Script flow | /10 | Hook -> Problem -> Solution -> CTA | | 8 | Disclosure | /10 | AI disclosure compliant with region (see variant) | | 9 | Platform fit | /10 | Correct aspect ratio, duration, format for platform | | 10 | CTA | /10 | Clear call-to-action, easy to execute | ### Action thresholds | Tier | Score | Action | |------|-------|--------| | **Excellent** | 90-100 | Publish now | | **Good** | 70-89 | Publish, note improvements for next round | | **Needs fix** | 50-69 | Fix items scoring under 7, then re-render | | **Redo** | <50 | Rebuild from script + voice + avatar | --- ## Output template ```markdown # AI Avatar Video — [Title] | [Region variant] | [Date] 1. Workflow used: [Single / Translate / Batch / Hybrid] 2. Script: [Content, 150-300 words] 3. Voice: [Tool] — [Voice ID / clone name] — Consent: [Yes / N/A] 4. Avatar: [Tool] — [Avatar ID / custom] 5. QA Score: [X]/100 (10 criteria) 6. Disclosure (per region variant): [Text + placement] 7. Publish: [Platform] — [Aspect ratio] — [Link] ``` --- ## Quality checklist - [ ] Information collection completed (4 questions) - [ ] Tier picked (Free / Pro / Enterprise) and aligns with budget + volume - [ ] Workflow picked (Single / Translate / Batch / Hybrid) - [ ] Voice clone consent recorded (if cloning a real person) - [ ] Avatar setup checklist completed before recording - [ ] Anti-detection techniques applied for the target platform - [ ] Region variant read and disclosure compliant - [ ] QA Score >= 70 before publishing --- ## Related skills - `25-voice-clone-podcast-global` — voice clone deep-dive + podcast pipeline - `04-script-video-global` — script writing for AI avatar - `26-thought-leadership-content-global` — content strategy for personal brand - `references/ai-video-disclosure-global` — full legal reference - `references/voice-clone-prompts-global` — voice clone training prompts --- *Global Skill 24 (AI Avatar Production) | Over Powers Agency | v1.1.0* --- REFERENCE FILE: skills/en/references/voice-clone-prompts-global.md # Voice Clone Prompts — Global Reference > **Reference file** — Load when using `23-voice-clone-pb-global`, `22-ai-avatar-pb-global`, or any personal brand skill that involves voice cloning. > Scope: international (US, UK, AU, SEA, LATAM English speakers). Targets ElevenLabs, HeyGen, Murf, PlayHT, Resemble.ai. --- ## 1. Recording Best Practices Quality of the cloned voice depends almost entirely on the quality of your input sample. A 30-second clip from a quiet professional studio outperforms a 30-minute clip recorded in a noisy cafe. Optimize the source. --- ## 2. Recording Principles ### Environment - **Quiet room:** No HVAC, fans, fridge hum, or traffic noise. Test by recording 10 seconds of silence — if you can hear anything, fix it before continuing. - **Acoustic treatment:** Soft surfaces absorb reflections. Record in a room with carpet, curtains, bookshelves, or sofas. Avoid bathrooms, kitchens, and empty rooms with hard walls. - **Time of day:** Early morning or late evening usually has less ambient noise. - **Phone silenced:** Even on silent, vibrations from notifications can register on sensitive microphones. ### Microphone - **Best — broadcast condenser:** Shure SM7B (XLR, ~USD 399), Rode NT1 (XLR, ~USD 269), Neumann TLM 103 (premium tier) - **Good — quality USB:** Blue Yeti (~USD 130), Rode NT-USB+ (~USD 169), Shure MV7 (USB + XLR, ~USD 249) - **Acceptable for testing:** Lavalier mic with smartphone (Rode Wireless GO, ~USD 199 set), good wired earbuds with mic - **Avoid:** Laptop built-in microphone, gaming headsets, cheap USB mics under USD 30 ### Distance and angle - **Distance:** 6-12 inches (15-30 cm) from your mouth — about a hand's width - **Angle:** Microphone directly facing your mouth, tilted 15 degrees off-axis to reduce plosive air bursts on "P" and "B" sounds - **Pop filter:** Recommended — reduces breath pops without changing tone. USD 15-30 investment. - **Posture:** Sit upright with shoulders relaxed. Slouching constrains your diaphragm and changes voice timbre. ### Length and content - **Minimum:** 3 minutes of clean audio for most cloning tools (1 minute possible on some platforms but reduces fidelity) - **Recommended:** 5-10 minutes including varied emotional range - **Premium voice tiers (ElevenLabs Professional Voice Clone):** 30 minutes to 3 hours for maximum fidelity - **Read from a prepared script** — speaking naturally introduces "um," "uh," pauses, and inconsistent pacing that degrade the clone - **Tip:** Print script on paper or set it at eye level. Looking down at your phone changes vocal projection and posture. --- ## 3. Three Sample Scripts By Accent > Each script is ~120 words and covers diverse phonemes. Read at conversational pace, pausing ~1 second between paragraphs. Maintain natural intonation — do not flatten your voice into a "podcast voice" that does not match your normal speech. ### Sample A — US English (General American) — Business Topic ``` Hello and thank you for tuning in. Today I want to share some thoughts on building a personal brand in the modern marketing landscape. When I started, I had nothing but a laptop, a strong work ethic, and a willingness to learn from every mistake. The market has changed dramatically over the past five years, and the people who adapt fastest will lead the next decade. What matters most is understanding your audience. What problems keep them up at night? What outcomes do they actually want? Once you answer those questions honestly, your content writes itself. I have tested dozens of approaches, from short-form video to long-form newsletters, and consistency wins every time. Thanks for listening, and I will see you in the next episode. ``` ### Sample B — UK English (Received Pronunciation reference) — Business Topic ``` Good day, and welcome to today's discussion. I want to share some observations on personal branding for professionals navigating today's marketing environment. When I began, I had little more than a laptop and a determination to keep iterating. The marketplace is more competitive than ever, and those who embrace new tools thoughtfully will pull ahead. The fundamental question is always the same: what does your audience genuinely need? Not what you want to tell them, but what would make a real difference in their work or their lives. Once you can answer that question honestly, the creative direction becomes obvious. I have experimented with many formats over the years. Consistency, clarity, and respect for the audience remain the constants. Thank you for joining me. ``` ### Sample C — Australian English — Business Topic ``` G'day everyone, and welcome to today's chat. I want to talk about something that comes up in nearly every conversation I have with founders — building a personal brand without losing your authentic voice. When I started, I had a laptop, an idea, and zero followers. Fast-forward a few years, and the principles that actually work are surprisingly simple. First, know who you are speaking to. Not a vague audience persona — a real person with real problems. Second, show up consistently, even when no one is watching. Third, share what you learn as you learn it, mistakes included. Polished content has its place, but raw honesty is what builds trust. Cheers for tuning in, and I will see you next week. ``` ### Sample D — Singapore / SEA International English — Business Topic ``` Hello and welcome. Today I want to share some practical observations on personal branding for professionals across Southeast Asia and the broader international market. When I started building my brand, I worked with what I had — a laptop, a network of mentors, and a commitment to publish consistently every week. The market here is unique. We sit at the intersection of multiple cultures, multiple languages, and rapidly changing consumer preferences. What works in one country does not always translate to another, and that creates both a challenge and an opportunity. The professionals who succeed are those who respect local context while building a globally relevant brand. Thank you for listening, and I look forward to our next conversation. ``` --- ## 4. Emotion Range Script > Purpose: train the AI to handle different emotional registers. Read all four segments back-to-back with a 3-second pause between segments. Each segment is approximately 30 seconds. ### Segment 1 — Neutral (introduction) ~30s ``` Hello, my name is [Your Name]. I work in marketing and digital communications. Every day, I help small and medium businesses build their brand presence across social media, search, and email. The work requires patience, creativity, and the ability to analyze data without losing sight of the human story behind the numbers. ``` ### Segment 2 — Positive / Happy (sharing a win) ~30s ``` And I am genuinely excited to share something with you today. Last month, our team's campaign exceeded its target by forty percent. The team did extraordinary work. The client is thrilled, revenue is up, and most importantly, we proved that our approach actually works at scale. This is the kind of result that reminds me why I love what I do. ``` ### Segment 3 — Reflective (lesson from failure) ~30s ``` But it has not always gone smoothly. I once led a project that failed publicly and painfully. At the time, I was convinced I had done everything right. The lesson, looking back, was this: success rarely comes from working harder. It comes from working on the right things. Sometimes the best decision is to stop, step back, and look at the full picture. ``` ### Segment 4 — Decisive / CTA ~30s ``` So if you have been waiting for the right moment to start, let me be direct: start today. Not next month. Not when conditions feel perfect. Not when you finally feel ready. Every day you delay, your competitors get further ahead. Click the link below, sign up now, and let's build your brand together. The right time was yesterday. The next best time is right now. ``` --- ## 5. Tool-Specific Prompts and Settings ### ElevenLabs Voice Design — 6 Voice Styles ElevenLabs Voice Lab uses three core parameters: **Stability** (consistency), **Similarity** (closeness to source), and **Style** (expressiveness). | Style | Stability | Similarity | Style | Best For | |-------|-----------|------------|-------|----------| | **Professional** | 0.75 | 0.80 | 0.30 | Corporate training, webinars, B2B explainer videos | | **Warm** | 0.50 | 0.85 | 0.50 | Podcasts, personal stories, audiobook narration | | **Energetic** | 0.35 | 0.75 | 0.70 | Short-form video, ads, motivational content | | **Calm** | 0.80 | 0.85 | 0.20 | Meditation, ASMR, sleep stories, slow-paced docs | | **Authoritative** | 0.70 | 0.80 | 0.40 | Thought leadership, keynote talks, expert commentary | | **Friendly** | 0.45 | 0.80 | 0.55 | Community videos, Q&A, vlog-style content | **Tips:** - Start with Similarity 0.80; raise to 0.90+ for closer match to source - Lower Stability for more emotional range (TikTok, Reels) - Higher Stability for corporate and long-form - Test with 3-5 different sentences before producing batch content - Save voice presets per use case rather than tweaking each time ### HeyGen Custom Voice 1. Go to **Voice Lab** > **Create Custom Voice** 2. Upload WAV or MP3 file (minimum 1 minute, recommended 3-5 minutes for standard voice; 30+ minutes for Professional) 3. Name the voice descriptively (example: "John_Business_Tone" or "Sarah_Friendly_US") 4. Select language and accent (English-US, English-UK, English-AU, English-Global, plus 40+ other languages) 5. Allow 5-15 minutes for processing 6. Test with 2-3 sample sentences before producing video 7. **Tip:** Upload multiple voice profiles for different use cases (business, casual, energetic) rather than relying on a single voice to do everything ### Murf Murf offers both voice cloning (paid tier) and a library of pre-built voices. For cloning: 1. Navigate to **Voice Cloning** in your Murf workspace 2. Upload sample audio (5 minutes minimum recommended) 3. Choose voice profile: Conversational, Promotional, Newscast, Inspirational, Storytelling, or Documentary 4. Voice settings to adjust: - **Pitch:** -50 to +50 (default 0) - **Speed:** 0.5x to 2.0x (recommended 0.9-1.1x for personal brand) - **Emphasis:** Mark specific words for vocal stress - **Pauses:** Insert custom pauses in milliseconds ### PlayHT PlayHT offers Voice Cloning with both Instant Clone (30 seconds) and High-Fidelity Clone (5+ minutes). 1. Go to **Voice Library** > **Clone Your Voice** 2. Choose Instant Clone for quick tests or High-Fidelity for production 3. Upload audio (recommended: WAV at 44.1kHz) 4. Configure generation settings: - **Voice Engine:** PlayDialog (most natural for conversational), PlayHT 2.0 Turbo (faster) - **Temperature:** 0.5 (consistent) to 1.5 (expressive) — start at 0.8 - **Quality:** Draft for testing, Premium for final output ### Resemble.ai Resemble specializes in real-time and emotional voice synthesis. 1. **Build > Voices > New Voice > Rapid Voice Clone** (3 minutes) or **Professional Voice Clone** (10+ minutes) 2. Upload audio matched to a script Resemble provides for higher accuracy 3. After clone is built, generate audio with emotion tags inline: - `This is exciting news!` - `I wish things had gone differently.` - `This is unacceptable.` - `Today's report covers three topics.` 4. Adjust pitch, speaking rate, and emphasis per phrase using the Resemble editor --- ## 6. Common Pitfalls — 5 Typical Issues ### Issue 1: Background noise contaminates the clone - **Symptom:** Generated voice has subtle hiss, hum, or echo - **Fix:** Re-record in a quieter environment. Use Audacity (free) Noise Reduction effect on existing samples — capture 2 seconds of room tone, use as noise profile, apply at 12 dB reduction - **Prevention:** Test record 10 seconds and play back through headphones before recording the full script ### Issue 2: Speaking too fast - **Symptom:** Clone "swallows" syllables, pronunciation muddled - **Fix:** Slow down 20% from your natural rate. Pause clearly between sentences - **Prevention:** Set a metronome to 100-110 BPM and pace your reading. Or read with a slight smile in your voice — it slows you down naturally ### Issue 3: Sample too short - **Symptom:** Clone is robotic, lacks the texture of your real voice - **Fix:** Provide at least 3 minutes; ideal is 5-10 minutes with varied content - **Prevention:** Combine Sample A/B/C/D (Section 3) + Emotion Range Script (Section 4) for ~6-8 minutes of varied training data ### Issue 4: Monotone reading - **Symptom:** Clone sounds flat, lifeless, or sleep-inducing - **Fix:** Include the Emotion Range Script (Section 4). Emphasize key words. Use pitch variation as you would when telling a story to a friend - **Prevention:** Imagine you are speaking to one specific person you care about, not "reading aloud" ### Issue 5: Mixing accents or registers in the same sample - **Symptom:** Clone is inconsistent, switches register mid-sentence - **Fix:** Choose one accent and one register, maintain throughout the sample - **Prevention:** Use only the script matching your primary accent (Section 3). Do not mix US, UK, AU, or SEA samples in a single voice profile --- ## 7. Voice Consistency Check After cloning, test with 5 **new** sentences not present in the training sample to evaluate quality. ### 5 Test Sentences ``` 1. "Today is a great day to start a new project." 2. "Why do you think that? I completely disagree with that view." 3. "Wow, this result is far beyond what I expected!" 4. "Please stay calm and think carefully before making this decision." 5. "Sign up today to get a 50 percent discount available only for the next 24 hours." ``` ### Scoring Rubric (0-2 per criterion, 10 total) | Criterion | 0 | 1 | 2 | |-----------|---|---|---| | **Tone match** | Wrong tone | Approximate | Accurate | | **Speed match** | Too fast or slow | Slightly off | Natural pace | | **Emotion range** | Monotone | Limited expression | Matches intended emotion | | **Pronunciation** | Multiple errors | Minor errors | Accurate (especially proper nouns, numbers, jargon) | | **Natural feel** | Robotic | Acceptable | Sounds like a real person | ### Overall Assessment | Total | Grade | Action | |-------|-------|--------| | 8-10 | Excellent | Use in production. Re-review quarterly. | | 6-7 | Good | Acceptable. Tune Stability and Similarity settings. | | 4-5 | Re-record | Sample needs improvement. Check mic, environment, technique. | | 0-3 | Start over | Major issues. Re-read Section 2 and re-record from scratch. | --- REFERENCE FILE: skills/en/references/ai-video-disclosure-global.md # AI Video Disclosure — Global Reference > **Disclaimer:** This document is for general informational purposes only. It is **NOT legal advice**. AI disclosure laws, FTC/ASA guidance, and platform policies evolve rapidly and vary by jurisdiction, industry, and business model. Always consult a qualified attorney licensed in the relevant jurisdiction before launching campaigns that use AI-generated voice, likeness, or content. > **Reference file** — Load when using `24-ai-avatar-production-global`, `25-voice-clone-podcast-global`, `26-thought-leadership-content-global`, or any personal brand skill that produces AI-generated content. > Last updated: 2026-10. Review cadence: every 6 months. Statutory dates and instrument numbers: verify on the primary source (EUR-Lex, ftc.gov) before relying on them. --- ## 1. Legal Framework By Region ### United States — FTC Endorsement Guides (16 CFR Part 255) and Consumer Reviews Rule (16 CFR Part 465) Two FTC instruments matter here: the **Endorsement Guides** (16 CFR Part 255) and the **Rule on the Use of Consumer Reviews and Testimonials** (16 CFR Part 465). Part 255 (Endorsement Guides) is where the duty to **disclose any material connection** between an endorser and the advertiser comes from. Part 465 (consumer reviews rule) addresses fake reviews, including AI-generated ones, undisclosed insider reviews and bought reviews. Check current text and FTC guidance on ftc.gov before relying on the details below. - **Material connection** = anything that could affect the weight of the endorsement (payment, free product, employment, family relationship, AI-generated likeness of a paid spokesperson, etc.) - **Clear and conspicuous** standard: a reasonable consumer must notice the disclosure without effort - **Fake reviews and testimonials:** fabricated testimonials, including AI-generated ones that imply real customer experience, are deceptive practices under Section 5 of the FTC Act and fall under the Part 465 consumer reviews rule (fake or AI-generated reviews, undisclosed insider reviews, bought reviews). Disclosure of a paid or free-product relationship is a Part 255 duty, not a Part 465 one. Penalties: see the FTC — verify on ftc.gov. - **In-stream disclosure** is the safe practice for sponsored content on TikTok, YouTube, Instagram, etc. - **Influencer disclosure:** #ad, #sponsored, "Paid partnership with [Brand]" — placed where the viewer cannot miss it **Resource:** FTC.gov "Disclosures 101 for Social Media Influencers" (verify on ftc.gov) ### European Union — AI Act Article 50 (applies from 2026-08-02) The EU AI Act (Regulation (EU) 2024/1689) entered into force on 2024-08-01. Article 50 addresses transparency obligations and **applies from 2026-08-02** (verify on EUR-Lex). - **Article 50(2):** Providers of generative AI systems must ensure outputs are **marked as artificially generated** in a machine-readable format (e.g., watermark, metadata). Systems already on the market before 2026-08-02 reportedly have until **2026-12-02** (reported; verify on EUR-Lex — sources differ on scope). - **Article 50(4):** Deployers of AI systems that generate or manipulate image, audio, or video content constituting **deepfakes must disclose** that the content has been artificially generated - **Exception:** Authorized criminal investigations; artistic/satirical works (but still must disclose in a manner that does not hamper enjoyment) - **Scope:** Applies to anyone placing AI systems on the EU market or whose output is used in the EU, regardless of where the provider is established - **Code of Practice on marking and labelling AI-generated content:** published mid-2026, voluntary. Following it is a compliance aid, not a safe harbour — verify on EUR-Lex / the Commission site. - **Penalties:** see Art. 99 — verify on EUR-Lex - **Disclosure standard:** deepfake deployers must disclose no later than the time of first interaction or exposure, in a clear and distinguishable manner ### European Union — Unfair Commercial Practices Directive (UCPD, 2005/29/EC) Independent of the AI Act, the UCPD prohibits misleading commercial practices across all 27 EU member states. - **Article 6:** Misleading actions — false information about main characteristics, results, endorsements - **Article 7:** Misleading omissions — failing to disclose material information including the commercial intent - **Annex I (blacklist):** Always prohibited practices including "Falsely claiming or creating the impression that the trader is not acting for purposes relating to his trade" - **AI implications:** Undisclosed AI-generated testimonials, fake AI influencers presented as real, or AI avatars without disclosure of artificial nature can fall under misleading practices ### European Union — GDPR (voice and face data) Under GDPR, biometric data including voice patterns and facial features qualify as **personal data** (and **special category data** under Article 9 when used for unique identification). - **Voice cloning consent:** Cloning a real person's voice requires lawful basis under Article 6, typically explicit consent (Article 9(2)(a) for biometric identification) - **Face/likeness:** Same standard — written consent recommended, with the right to withdraw at any time - **Documentation:** Maintain a consent log including scope (which content, which campaigns, which platforms), duration, withdrawal mechanism - **Cross-border transfer:** If AI tool processes data outside EU/EEA, ensure SCCs or adequacy decision applies - **Data Protection Impact Assessment (DPIA):** Likely required for systematic voice/face cloning operations ### Southeast Asia | Country | Body | Key Rule | |---------|------|----------| | **Singapore** | ASAS (Advertising Standards Authority of Singapore) | 2024 guidelines: clearly disclose AI-generated endorsements; PDPA applies to voice/face data | | **Indonesia** | AKARI / KPI | Broadcasting law requires disclosure of digitally manipulated content in commercial broadcasts | | **Malaysia** | MCMC + ASA Malaysia | Content Code requires honesty; AI-generated content in ads should be disclosed | | **Thailand** | NBTC + Consumer Protection Act | PDPA (similar to GDPR) governs voice/face data; misleading ads prohibited | | **Philippines** | DTI + Data Privacy Act | Consumer Act prohibits deceptive sales practices; AI disclosure recommended | | **Vietnam** | See the VN legal hub | AI-labelling duties: Law 134/2025/QH15 + Decree 142/2026/ND-CP. Details: `skills/vi/references/phap-ly-marketing-vn.md` section 4 (verify on the official gazette) | ### Latin America | Country | Body | Key Rule | |---------|------|----------| | **Brazil** | CONAR (Conselho Nacional de Autorregulamentacao Publicitaria) | 2024 guidelines on AI in advertising require identification; LGPD governs voice/face data | | **Mexico** | PROFECO + IFT | Consumer protection law prohibits misleading ads; emerging AI disclosure norms | | **Argentina** | AAIP + CONARP | AAIP enforces data protection; CONARP recommends disclosure of AI-generated endorsements | | **Chile** | SERNAC | Consumer rights law applies; new data protection law (effective 2026) aligns with GDPR | | **Colombia** | SIC | Habeas Data law governs biometric data; misleading advertising prohibited | ### Other Major Markets - **UK:** ASA CAP Code rules 3.1 (misleading) and 3.7 (material information) — AI ads must be clearly identifiable. UK GDPR mirrors EU GDPR for biometric data. - **Canada:** Competition Act prohibits false/misleading representations; PIPEDA covers biometric data; CRTC oversees broadcasting - **Australia:** ACCC enforces Australian Consumer Law (false/misleading conduct); Privacy Act 1988 covers biometric information - **India:** ASCI guidelines + Consumer Protection Act 2019 + DPDP Act 2023 (biometric data) --- ## 2. Three-Tier Disclosure Rules ### MANDATORY (always disclose) - **Paid advertising** using AI avatar, AI voice, or AI-generated likeness - **Sponsored content** where a creator received compensation, free products, or other consideration - **Affiliate marketing** with revenue share, commissions, or referral fees - **Brand partnerships** of any form (long-term contracts, gifting, "ambassador" programs) - **Comparative advertising** featuring AI representations of competitors or their products - **Health, financial, or legal claims** delivered via AI avatar (additional disclaimers required) ### RECOMMENDED (best practice) - **Organic content** with any material connection (your own product, employer, investments) - **Educational/informational content** on a topic where you have undisclosed financial interest - **Personal brand content** building thought leadership when using AI to scale production - **B2B content** where audience may assume direct human delivery - **Long-form content** (podcasts, courses, webinars) where AI generation could be missed ### OPTIONAL (use judgment) - **Pure entertainment / parody** clearly labeled as such (e.g., satire account) - **Internal team training** not distributed externally - **Concept tests and A/B drafts** never published - **Private communications** within a defined small group with prior disclosure - **Artistic works** clearly framed as creative AI exploration --- ## 3. When To NEVER Use AI Avatar ### Impersonation of real people without consent Violates: defamation law, right of publicity (US states), Article 8 ECHR (privacy), GDPR (biometric data), copyright on voice/likeness in many jurisdictions. **Risk:** Civil damages, criminal fraud charges, permanent platform bans. ### Political deepfakes Many jurisdictions (US states like Texas, California; UK Online Safety Act) now prohibit or heavily restrict synthetic political content. In the EU, the relevant layers are the AI Act Art. 50(4) deepfake disclosure (applies from 2026-08-02), Regulation (EU) 2024/900 on political advertising, and national election rules — verify on EUR-Lex. **Risk:** Criminal penalties, election law violations, platform-wide bans, civil liability. ### Fake testimonials and reviews Violates: FTC Act Section 5, EU UCPD Annex I, consumer protection acts in nearly every jurisdiction. **Risk:** FTC civil penalties (verify on ftc.gov); national enforcement under UCPD; class action exposure. ### Unauthorized medical, health, or financial advice Beyond AI disclosure, these often require licensed professional involvement. **Risk:** Regulatory action by FDA, FTC, SEC, MHRA, EFSA, or local equivalents; possible criminal liability for unlicensed practice. ### Content featuring minors without verifiable guardian consent GDPR Article 8, COPPA (US), Online Safety Act (UK), and similar laws impose elevated protections. --- ## 4. Per-Platform Disclosure Templates ### TikTok - **In-video text overlay** (first 3 seconds, contrasting color, minimum 24px equivalent): "AI-generated voice and avatar" - **Caption disclosure** (first line, before "more"): "This video uses AI avatar technology. #AIgenerated #SyntheticMedia" - **TikTok AI-generated content label:** Use the built-in toggle in Post Settings when content is significantly altered or fully AI-generated - **Profile bio note** (optional but recommended for AI-heavy accounts): "Some videos use AI avatar. Authentic content by [Your Name]" ### YouTube - **Description box** (first paragraph, before fold): ``` Note: This video was created with AI avatar technology. The voice and likeness are based on the creator's verified consent. All content is reviewed and approved by [Your Name] before publication. ``` - **YouTube Studio "Altered content" disclosure:** Toggle in Upload > Details for realistic synthetic or significantly altered content (platform policy — check YouTube Help Center, 2026-10) - **End screen note** (last 5 seconds): "AI-generated avatar | Content by [Your Name]" - **Community Guidelines label:** Use the AI-generated content tag in YouTube's metadata ### Meta — Facebook and Instagram - **Ad disclosure** (required for paid ads): Include disclosure in primary text within the first 125 characters ``` Ad uses AI-generated voice and avatar. Content reflects [Brand]'s actual products and services. ``` - **Post caption** (organic, last paragraph): ``` This content uses AI avatar technology. Voice and likeness based on [Name]'s consent. ``` - **Story sticker:** Use the "Made with AI" text sticker - **Reels:** Add in-video text overlay matching TikTok format - **Meta AI-generated label:** Use the platform's built-in labeling when uploading clearly synthetic content ### LinkedIn - **Post note** (final paragraph in italics): ``` *Note: This video uses AI avatar technology to streamline content production. The opinions and analysis are mine.* ``` - **Featured section / About:** Add a line in your profile About section if AI content represents a substantial portion of your feed ### Personal Website / Blog - **Footer disclaimer** on every page with AI content: ``` Some content on this site is produced with AI avatar and voice technology. All content is reviewed and approved by [Your Name] prior to publication. ``` - **About page** dedicated section: ``` About this site's technology: I use AI avatar and AI voice as production tools. All editorial decisions, opinions, and content direction are mine. AI assists with delivery format only. ``` ### Podcast Platforms (Spotify, Apple Podcasts) - **Show description:** Note in the standing description if voice is AI-generated - **Episode notes** (first line): "This episode uses AI voice technology. Authentic script by [Host Name]." --- ## 5. Five Disclosure Copy Templates ### Template 1 — Short (overlay, story, character-limited) ``` AI-generated voice and avatar ``` ### Template 2 — Medium (post caption, video description) ``` This content uses AI Avatar technology. Voice and likeness based on [Name]'s consent. Content authored and reviewed by [Name]. ``` ### Template 3 — Long (paid advertising, compliance-heavy industries) ``` NOTICE: This advertisement uses AI Avatar and AI Voice technology to generate the visual presentation. The depicted likeness is based on actual photography of [Name] used with consent. All claims, product information, and recommendations reflect the actual products and services of [Brand]. Content has been reviewed and approved by [Name] prior to publication. ``` ### Template 4 — English standard (international audiences) ``` This content was created with AI avatar technology. Voice and likeness are based on a real person with verified consent. All content is reviewed and approved by the creator prior to publication. ``` ### Template 5 — Bilingual (EN + local language note) ``` This video uses AI technology | Authentic content by [Your Name] [Local language repetition of the same message] ``` Example for Spanish: `This video uses AI technology | Este video utiliza tecnologia de IA | Contenido autentico por [Nombre]` --- ## 6. Decision Tree — "Do I Need To Disclose?" ```mermaid flowchart TD A[Does your content use AI avatar, voice, or generated likeness?] -->|No| B[No disclosure needed] A -->|Yes| C{Is the content commercial in nature?} C -->|Yes — paid ad, sponsored, affiliate, brand deal| D[MANDATORY disclosure] C -->|No| E{Is the content public-facing?} E -->|Yes — social media, website, podcast| F{Do you have any material connection to the topic?} E -->|No — internal, private, draft| G[OPTIONAL — internal note recommended] F -->|Yes — own product, employer, investments| H[MANDATORY under FTC / UCPD] F -->|No — purely informational| I{Are you in EU or targeting EU users?} I -->|Yes| J[MANDATORY under EU AI Act Art. 50] I -->|No| K[RECOMMENDED — best practice] D --> L[Use Template 3 / platform-specific format from Section 4] H --> L J --> M[Use Template 2 or 3 with AI Act language] K --> N[Use Template 1 or 2] G --> O[Internal team note only] ``` --- ## 7. Update Log - **2026-Q1:** Initial global reference released. - **2026-10:** EU AI Act Art. 50 application date (2026-08-02), reported Art. 50(2) grace date, and voluntary Code of Practice added; FTC Part 465 named; penalty figures removed (point to Art. 99 / ftc.gov); Vietnam row now points to the VN legal hub. - **Review cadence:** Every 6 months or upon major regulatory change (EU AI Act milestones, FTC final rules, major national AI laws). --- Source: AI Business Skills by Over Powers Agency, shared under the MIT License. MIT License Copyright (c) 2025 Over Powers Agency Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.