Viral video script

Viral Video Script

Write the words for a short-form video (reel / TikTok / Short) so the human brain can't scroll past it. The formula is distilled from Tuan Le's talk "What Getting 3 Billion Views Taught Me About Human Psychology" — six psychological principles plus a story skeleton. It works for "boring" products (instant ramen, SaaS, phone plans) precisely because it sells how brains work, not the product.

The deliverable is a script, not a video. Its outputs drop straight into this repo's pipeline: the clean narration track feeds voice-clone / avatar-reel-composer.

The formula (six principles)

Apply ALL six. They are not a menu.

1. Familiarity beats originality (mere-exposure effect). Don't invent a format — steal a proven one (reaction, challenge, "walk up to a stranger", office tour, "how much do you pay for rent", "hey chef, make me something"). The brain watches what it already recognizes and has liked before. Originality reads as confusion, and confusion is the fastest path to a scroll.

2. The curiosity gap hook. The first ~2 seconds open a gap between what the viewer knows and what they want to know — an itch they can't leave unscratched. Never lead with the product; the instant a brain tags content as an ad, it scrolls. Trigger curiosity, disbelief, or identity recognition instead. The product appears naturally once they're already watching.

3. Sell identity, not the product (Means-End Chain). Every product has three layers: (1) attributes = what it is, (2) functional consequences = what it does, (3) psychological value = what it means to the viewer's identity/emotions. Most brands talk layer 1. Talk layer 3. ("Supplement" → healthier body → energy & mood → "I feel in control of my body and my life.") When you speak to identity, the viewer feels understood, not sold to — and the brain stops flagging the ad.

4. The credential shortcut (authority bias). The brain takes shortcuts; the biggest is authority. Put a credential in the first ~2 seconds ("Thai chef", "Harvard student", "engineer who sold his company", a real kitchen + chef's outfit). It buys instant trust before the viewer evaluates anything.

5. Build in shareability (social currency). People don't share what they like — they share what makes them look good (smart, funny, in-the-know) to their friends. You are the wingman: make the sharer look like they have great taste. The premise must be explainable in one sentence; if they can't describe it fast, they won't share it. Drivers that spread: humor, surprise, awe. (Sadness and anger get views but don't spread the same way.)

6. The story skeleton (narrative transportation). Inside a story, critical thinking slows and emotion takes over — the best state for the brand to appear. Every script follows: Hook → Problem → Story → Payoff. Editing rules: cut any frame that isn't delivering new info, keep clips a few seconds each (every cut resets the brain's attention clock), always burn in captions (dual audio+visual processing boosts retention), zero dead space.

North star: the product is almost irrelevant. What matters is whether you understand the person watching — what catches their attention and what makes them share. Put people first.

Inputs to gather (ask only what's missing)

Input	Needed for	Example
Product / brand / topic	the subject	"Buldak instant ramen"
Audience	identity layer (#3), tone	"spice-loving Gen Z snackers"
Platform	pacing / length	tiktok | reels | shorts
Language	the spoken copy	es | en (default: match the user)
Target length	beat budget	~30s (default 25–40s)
Credential available	the authority hook (#4)	"Korean street-food chef"
A format to steal (optional)	#1 — else propose 2–3	"spicy-challenge reaction"

If a credential or stealable format is missing, propose options rather than skipping the principle.

Workflow

- [ ] 1. Gather inputs (above). Propose a format-to-steal + credential if absent.
- [ ] 2. Scaffold the beat sheet JSON.
- [ ] 3. Fill every field, writing the spoken copy beat-by-beat in the formula's voice.
- [ ] 4. Validate against the formula. Fix every FAIL, weigh each WARN.
- [ ] 5. Render the shooting script (.md) + narration track (.txt).
- [ ] 6. Hand off (read aloud / send to voice-clone / avatar-reel-composer).

Step 2 — scaffold (writes <out>/<slug>.script.json pre-structured with the formula):

python3 .cursor/skills/viral-video-script/scripts/scaffold.py buldak-spicy \
  --product "Buldak instant ramen" --language en --platform tiktok --seconds 32 \
  --out scripts_out/

Step 3 — write. Edit the JSON. The hard rules the validator enforces:

• format_steal, credential, identity_value, shareable_premise are all non-empty.
• beats contains hook → problem → story → payoff in that order.
• hook.spoken is ≤ ~14 words (≈2s) and must not contain the product name.
• shareable_premise is ONE sentence, ≤ ~20 words.
• share_trigger ∈ {humor, surprise, awe}.
• captions: true, cut_every_seconds ≤ 4, total spoken copy fits target_seconds.

Step 4 — validate (re-run until it passes; treat WARN as review notes):

python3 .cursor/skills/viral-video-script/scripts/check_script.py scripts_out/buldak-spicy.script.json

Step 5 — render (writes <slug>.script.md + <slug>.narration.txt next to the JSON):

python3 .cursor/skills/viral-video-script/scripts/render_script.py scripts_out/buldak-spicy.script.json

Writing the spoken copy (per beat)

• Hook (≈2s): the curiosity gap + the credential, NOT the product. Use the stolen format's recognizable opener. One short line.
• Problem (≈4–6s): the tension the viewer already feels — framed at the identity layer, not the product layer.
• Story (≈15–25s): the recognizable format plays out; the product enters naturally as part of the story, never as a pitch.
• Payoff (≈5–8s): resolve the gap + the identity reward, then the lightest possible product/CTA. End on the feeling, not the feature.

Keep sentences short and spoken (this is read aloud / lip-synced). Mark on-screen action and caption text per beat; assume a cut every 2–4s.

Output files

<slug>.script.json   # the beat sheet (source of truth; edit this)
<slug>.script.md     # human shooting script: beats, timing, VO, on-screen, captions, notes
<slug>.narration.txt # clean spoken VO only — feed to voice-clone / narrate.py

Handoff into this repo's pipeline

The narration track is what the avatar speaks:

# clone/say in an avatar's voice, or compose a full reel from the script
python3 .cursor/skills/voice-clone/scripts/...           # TTS the narration.txt
python3 .cursor/skills/avatar-reel-composer/scripts/compose_reel.py <storyboard.json> --finish

For B-roll-only or found-footage shots referenced in the script, see broll-generator and broll-finder.

Anti-patterns

1. Leading with the product in the hook (instant ad-flag → scroll). #2.
2. Inventing a format instead of stealing a proven one. #1.
3. Selling features (layer 1) instead of identity (layer 3). #3.
4. A premise you can't say in one sentence → unshareable. #5.
5. Relying on sadness/anger for reach — gets views, doesn't spread. #5.
6. Long takes, no captions, dead space → the brain tunes out. #6.

Additional resources

• The psychology behind each principle + the original case studies: REFERENCE.md
• Full worked example scripts (EN + ES): examples.md