skills/ugc-website-video/references/saas-monologue.md

# Site monologue — hook → the site solves it → result, ending on an action

The creator is on camera the **whole video**. The story sells an **outcome and the experience of
using the SITE**, not a physical object: a real friction, the site removing it, a concrete result —
and it **ends on the creator doing something with the product**. The voice is the continuous spine
under everything, including where the screenshot cards overlay.

## Language — always English

**Write the monologue in English even when the brief is in another language.** Seedance lip-syncs and
renders English cleanly and mangles other languages. Talk to the user in their language; the spoken
monologue and every on-screen word stay English. Only switch if the user explicitly asks for the
video in another language.

## The hero is the site, never a physical product

Talk about USING the site — even for an e-commerce store: "I found the perfect pair on their site in
two minutes", "their reviews sold me", "checkout took one tap". Describe what is on the site, never an
object in the creator's hands **during the body**. No "I'm holding / wearing / opening the product"
mid-video — that belongs to `ugc-review-video`. The one exception is the closer action below.

## Structure — hook → body → closer

The creator never cuts away; the site appears as overlay cards during the body. Write the body so it
**names the useful sections in the order their cards will pop** (from the section map: product/hero →
features → reviews → pricing).

**The first body beat NAMES the site / brand** ("so I opened Meller…", "this is where [brand] comes
in") — the composite anchors the **hero card (logo / navbar)** to it, so the viewer sees what the site
IS before any proof.

| Part | Runtime | Voice |
|---|---|---|
| **Hook (first 1–2s)** | ~8-12% | ALWAYS a bold, specific claim, ≤8 words, with a hook token (number / price / % / time / contrast / named result). No greeting, no pointing intro. |
| **Body** | ~70% | Walk the useful sections in order, first beat names the site. One concrete line per section while its card is up — "800 reviews and it's a 4.9", "checkout's one tap", "the plan's ten bucks". Each beat anchors a card. |
| **Closer (last ~15%)** | ~15% | The result, concrete (a number, a time), CTA riding inside it — while the creator does an action with the product. |

- Skip filler (footer, logo strip) — no line, no card.
- The whole thing is one breath of momentum; no section feels like a new scene.

### Hook ≠ naming the site — two DIFFERENT beats

Do NOT merge them. The **hook** grabs attention: ≤8 words, a hook token, **no pointing site-intro**
(not "This is X" / "Meet X") — it may reference the brand only inside a claim. **Naming the site as
the thing you use** is the FIRST BODY beat, and the hero card anchors to that beat, not to the hook.

- **Hook:** "Sold out? Not on adidas." (claim + token, grabs attention)
- **First body beat:** "so I went straight to the adidas site…" ← hero card anchors here.

## The hook (first 1–2 seconds) — the most important rule

The first 1–2 seconds decide everything. The hook is ALWAYS a bold, specific claim delivered short,
punchy and confident. Mandatory in every video, for any site type.

### Hard limits (checkable)

- **≤ 8 words, ≤ ~2 seconds, one clause.** No setup, no second sentence before the payload.
- **A hook TOKEN in the first breath — REQUIRED:** a number, a price, a %, a unit of time, a contrast,
  or a named result. Without a token it is NOT a valid hook ("these are amazing" fails).
- **First word is hook content** — not filler.

### Banned openers

- Fillers: `OK / So / Wait / Um / Well / Like`.
- **Pointing intros with no stake** (the most common weak pattern — it just points at the picture):
  `These are / This is / Here's / Here is / Meet / I found / I want to show / Let me show / Say hello
  to`. If the hook would start this way, rewrite it around a token.

### Angle taxonomy — choose the angle that fits

Pick ONE angle and write your own line in it; the examples show the PRINCIPLE, never paste them
verbatim. Every angle still carries a hook token.

| Angle | Use when | Example template |
|---|---|---|
| **Price contrast** | cheaper than it looks | "$65 sunglasses that look like $300." |
| **Secret / reveal** | premium look for little money | "Everyone thinks these are designer." |
| **Social-proof number** | lots of sales / reviews | "3 million people bought these." |
| **"Stop X" conflict** | a category people overpay in | "Stop overpaying for sunglasses." |
| **Transformation / result** | SaaS, service, earnings | "This site saved me 20 hours a week." |
| **Speed** | fast result / delivery | "Found my perfect pair in 60 seconds." |

Peak facial energy right at the top (carry that into the opener clip). Keep it believable — a bold but
plausible claim, never scammy.

### Hook self-check (reject & rewrite BEFORE generating the first clip)

- [ ] ≤ 8 words?
- [ ] a hook token (number / price / % / time / contrast / named result) in the first breath?
- [ ] does NOT start with a banned pointing / filler opener?
- [ ] does NOT just restate what the picture already shows?

## Monologue self-check (reject & rewrite BEFORE generating)

- [ ] does it sound alive said out loud — or like a read ad? (reads-as-ad → rewrite)
- [ ] contractions present (`they're / it's / I'm / that's`)?
- [ ] at least one conversational connector (`ok so / honestly / no joke / …`)?
- [ ] sentence length AND pace vary — not an even rhythm?
- [ ] does the peak line PERFORM the emotion (a vowel-stretch or a CAPS spike on the payoff word)?

## Delivery — lively, varied rhythm (never a flat read)

Spoken and energetic in tone at a **natural, lively pace, ~2.4–2.7 words/sec — never crammed or
rushed**. Say it in the clip prompt too: "natural, lively spoken delivery with varied rhythm and
emphasis, never rushed". Energy comes from expression and emphasis, NOT from talking fast — a crammed
clip is the number-one cause of the "talks super fast" render.

**The main rule — this is what kills the robot voice: vary the speed WITHIN the take.** Move quicker
through throwaway / connective lines, slow down to LAND the key number, the price and the CTA. A
constant even speed — fast OR slow — reads as a robot; the rhythm must breathe.

## Spoken texture — write how people TALK (MANDATORY)

- **Contractions, always** — `they're / it's / I'm / that's / you'll`, never the written-out forms.
- **1–2 conversational connectors per monologue** — `ok so`, `honestly`, `no joke`, `and get this`,
  `here's the thing` — but **never as the first word of the hook**.
- **Vary sentence length** — clipped 2–4 word lines mixed with longer ones. A uniform rhythm is
  banned.
- **One live, imperfect beat per monologue** — a micro self-correction, a thinking-out-loud aside, or a
  reaction to yourself ("wait, that's actually insane"). One, not more.

## Performed dialogue — write the emotion INTO the words (MANDATORY)

Seedance under-renders flat prose — it acts what the WORDS perform, so bake the performance into the
text, not into stage directions. Use these sparingly, in service of the beat:

- **Vowel-stretch on the peak word** — usually the number / price / result the line lands: "it's SO
  good", "saved me TWENTY hours", "it's like TEN bucks".
- **1–2 CAPS volume-spikes per line, MAX** — the emphasis word only, never a whole line in caps.
- **One broken sentence at the peak** — a real-time reaction that fractures the line: "it's— ok wait.
  LOOK." One per monologue, at the single hardest beat.
- **A whisper→spike swing** — drop low on the setup, snap up on the payoff.
- **A flat→spike→settle arc across the clip** — build in, peak on the result, settle into the CTA.

**Keep it human-scale** — the peak is a real gasp, a breaking grin, a genuine "wait, what", NEVER
staged screaming or an infomercial voice. Convey emotion through the WORDS, not through pauses: one
engineered pause maximum, and extra pauses bloat and break the render.

## The closer — the creator does an action with the product

- **Physical product (e-commerce / a real object):** the creator may **pick it up / hold it up** in the
  last beat — "and it showed up two days later, look". Product in hand is allowed **here only**.
- **Non-physical (SaaS / service / app):** the creator **acts on the phone / device** — picks it up,
  taps, turns it to camera with a blank or turned-away screen — "so I just tapped go and it was done."

Write the closer line to motivate that action so the closer clip can render it.

## Beats → cards (one card per useful still)

Break the body into **many concrete beats — ideally one beat per captured screenshot** (the dashboard,
a review, the price, a feature…). Tag each beat with its section label and word count so the composite
anchors that still's card at the moment it is narrated. **More beats + more captured stills = more
cards on screen** — that is how the video shows more of the site, since per-card time and gaps stay
fixed. Give the sections that matter (product, proof) more beats, minor ones fewer.

## Word density (natural pace, ~2.4–2.7 words/sec)

≤10s ~22-26 words · 11-12s ~28-33 · 13-15s ~35-40 (**hard ceiling — do NOT exceed ~40 words in a 15s
clip; beyond that Seedance crams the words and talks too fast**). Aim for the MIDDLE of the range —
the ceiling is a cap, not a target. Budget the total across hook + body + closer with the
~8-12 / ~70 / ~15% split.

## Craft pass (anti-slop)

- **Every claim carries one concrete** — a number, a time, a named comparison. "Cut reporting from
  three hours to ten minutes" beats "saves time". Praise without a concrete gets cut.
- **Banned words** — `literally`, `obsessed`, `game-changer`, `holy grail`, `hits different`; corporate
  filler `elevate / seamless / effortless / synergy / leverage / unlock / supercharge / revolutionary`.
  A bold result claim IS allowed in the hook.
- **Body words describe the REAL card on screen** — talk about the reviews when the reviews card is up;
  never promise UI the capture does not show. Keeps voice and picture honest.
- **One engineered pause** max, in the hook or closer only — a silent beat, not an em dash.

## Voice gender & accent

Follow the user's creator gender if given, else pick from `audience` / `category` and lock it for the
whole video. Accent is **American English, never British** — enforced in the clip prompt.

## Output of this step

- The full monologue text (English), saved verbatim to `output/script.txt`.
- The **hook** line saved to `output/hook.txt` (the caption step reuses it for the top plate).
- The **hook** and **closer** segments with word counts; the closer notes the product action.
- The **body** as an ordered list of `{ section_label, words }` — one entry per captured still (~6-10),
  in card order — which is what the composite uses to anchor each overlay card.