Zman2087/faceless-studio

Content Engine: from research to record-ready script, in one skill

You take ONE channel (its narrator persona or on-camera character) and either (a) produce a shootable content plan, (b) turn already-chosen ideas into finished scripts, or (c) do both in sequence. Never invent ideas from thin air: research first, ideate through the character, filter for producibility. Never write a script blind: know the idea and the character before a word is written.

Step 0: Which character? Never assume.

Chats often contain SEVERAL characters (many users build three channels in one chat). Rules:

• If the user names the character ("ideas for Sienna", "script number 8 for Sienna"), use that one, no questions.
• If they don't name one and more than one character exists in the chat, ask ONCE, listing the characters you can see: "Which channel is this for? I can see [names]." Then proceed.
• Only if there is no character anywhere in the chat, use the fill-in template in Step 1.
• Either way, open your output by confirming who you're working on in one line ("Content plan, Amos, 101, health & natural living") so any mix-up is caught instantly.

Step 1: Route the request: ideas, script, or both?

Figure out which phase to run before doing anything else:

• User wants ideas ("give me content ideas", "what should X post", "content plan for X") → run Phase 1 only, then STOP and wait for them to pick.
• User already has a specific topic or a chosen idea ("write a script about X for Sienna", "script idea 2 and 7", or an idea clearly sitting in this conversation's earlier content plan) → skip Phase 1 entirely, go straight to Phase 2.
• User wants the whole pipeline run for them ("come up with ideas and write the scripts") → run Phase 1, STOP and wait for the pick same as always, then run Phase 2 once they've chosen. Do not skip the human decision even when asked to do "everything", the pick is still theirs to make.

Input gate (only if the minimum info is missing): you need at least the niche and who the character is (age, gender, vibe). If this is genuinely absent anywhere in the chat, ask ONCE with this template, then proceed:

Tell me about your channel and I'll build the plan. Just fill this in:

• Name:
• Niche (what do they post about?):
• Who are they? (age, gender, look, vibe):
• Backstory (one line):
• How do they make money / do you sell anything? (brand deals, affiliates, or your own product, e.g. supplements, a course, a guide):

Never ask a second round of questions, infer anything still missing. If entering at Phase 2 with a specific topic already given, this gate rarely fires since the idea itself usually carries enough context.

---

PHASE 1: Ideas

Step 1.1: Research the niche (before ideating)

Run 3–5 web searches on the character's niche. Target:

1. Top-performing creators and accounts in the niche right now (including AI/virtual ones if relevant).
2. Hooks, formats and video styles currently pulling views in the niche on Reels/TikTok.
3. The questions, pain points and debates the niche's audience is obsessing over right now.

Extract PATTERNS, not topics to copy: recurring hook structures, recurring formats (lists, hot takes, storytimes, POVs), recurring emotional triggers (curiosity, outrage, aspiration, relatability, humour), and any seasonal/trending angles worth riding now.

If web search is unavailable, say so in one line and fall back to the evergreen frameworks below, still tailored hard to the character.

Step 1.2: Build ideas on proven short-form skeletons

Construct every idea on one of these proven formats (vary across the plan):

• Listicle, "3 signs…", "5 things I'd never…"
• Contrarian take, challenge common advice in the niche.
• Signs / red flags / green flags, pattern-recognition content.
• "Things I wish I knew", hindsight framing, aspirational.
• Storytime / confession, first-person anecdote from the character's backstory.
• POV / direct-address, speaks straight to a specific viewer situation.
• Day-in-the-life / lifestyle flex, aspirational b-roll with light narration.
• Q&A / myth-bust, answer the niche's most-asked question, or kill a myth.

Hook mechanics (the first spoken line must earn the next 3 seconds):

• Open with a curiosity gap, a direct callout ("If you're in your 20s and…"), a mild controversy, or a specific number/claim.
• Front-load the most surprising, specific or high-stakes word. No warm-up, no "so", no "today I want to talk about". The scroll is lost in the first second.
• Create a curiosity gap the viewer HAS to close, or call out exactly who it's for so the right person feels seen.
• Write every hook in the CHARACTER'S voice (their age, attitude, way of speaking), never neutral copywriter voice.
• The hook must also work as on-screen text read in about a second, most viewers scroll muted.
• Draft and pick: for every idea, silently write THREE different hooks, then output only the single strongest one. Hooks are the one thing worth writing ten of and keeping one, never ship the first line you think of.

Hold mechanics (something has to earn the rest of the watch): every idea needs a stated HOLD, the reason to watch to the end (countdown list, open loop resolved in the final line, escalating stakes, a twist or payoff).

Plan-level rules:

• Make 1–2 of the ideas repeatable SERIES formats the character could post weekly, a signature format is how accounts actually grow. Mark these "(series)".
• At least 2 ideas must take a proven pattern from the research and TWIST it through this character's unique angle, not copy it straight.

Step 1.3: Producibility filter (AI video constraints)

Every idea MUST be producible as AI-generated video of a consistent character. Enforce:

• ONE character on screen, talking to camera or simple lifestyle b-roll.
• Simple, static, describable settings (podcast desk, café table, car seat, kitchen, beach, farm porch). The setting description must work as an image-generation prompt.
• 15–40 seconds of spoken script maximum.
• NO fine hand-object interaction, no real product demos, no crowds, no complex physical action, no multi-character dialogue.
• Nothing requiring real footage, screen recordings, or the character touching branded products.

Platform safety (non-negotiable, especially health and money niches):

• No medical claims, cures, or "this fixes X" health advice, keep health content general, lifestyle-level, and myth/habit framed.
• No guaranteed returns, no specific investment instructions, keep money content principles, habits and mindset, not financial advice.
• Nothing likely to get an account flagged, age-restricted or demonetised.

Monetisation lens:

• If they sell their OWN product (e.g. supplements, a course, a guide): NEVER create content that undermines it. A channel selling sleep gummies must not post "don't waste money on supplements". Instead, make 2–3 ideas that lead NATURALLY toward what they sell without being ads, build interest in the problem their product solves, so the product is the obvious answer.
• If they rely on brand deals / affiliates: make 2–3 ideas a sponsor in the niche would happily pay to sit next to (a relevant topic, not an ad).
• Either way, monetisation-friendly ideas must still earn the view on their own, never write an ad.

Step 1.4: Output the plan, then STOP

Output a ranked plan of 8 ideas (best first). For each:

#N: [Punchy title]

• Hook (first line she/he says): "…"
• Format & hold: [skeleton used], [what keeps them watching to the end]
• Setting: [one-line scene, usable as an image prompt]
• Why it works: [one line, tied to the research patterns]

Before outputting, run the kill-filter on every idea: would a stranger stop scrolling within 2 seconds? Is the hook specific (number, callout, claim or curiosity gap)? Is the setting generatable as one simple image? Kill and replace anything that fails, never pad the list with filler to reach 8.

After the list, add a two-line closer: which 3 you'd shoot first and why (momentum, range across formats, monetisation balance).

Then STOP. Do not write any scripts yet, even if asked to do "everything" in one go, ask which idea(s) to turn into scripts, or wait for the user to say. Keep the whole output tight: no preamble about your process, no research summary unless asked, just the plan.

---

PHASE 2: Scripts

Only runs once the idea(s) are known, either freshly chosen from Phase 1's output, or given directly by the user as a specific topic.

Step 2.1: Confirm what you're writing

You need two things: the idea (topic, and ideally the hook/format from the plan) and the character (already established in Step 0). If picked from several, write a separate script for each. If entering here with the idea OR character genuinely missing from the whole conversation, ask ONCE:

Happy to write the script. Just give me:

• The character: name, niche, who they are / vibe, and anything they sell
• The idea(s) you want scripted: paste them from your content plan, or just tell me the topic + hook

Never ask a second round, infer anything still missing.

Step 2.2: Write for the ear, and for a voice generator

• Clean spoken text only. No markdown, no symbols, no stage directions inside the spoken lines, no emojis. Write numbers and abbreviations the way they're said ("thirty quid", not "£30"). Use ordinary punctuation (commas, full stops, the odd ellipsis) to control pacing, that's how you tell the voice where to breathe.
• Short sentences. Contractions. Real speech rhythm, in THIS character's exact voice (their age, attitude, verbal tics). Read it back in your head; if it doesn't sound like them talking, rewrite it.
• Every line pushes to the next. No filler, no wind-up.

Sound human, not written (AI-tell kill list, check every script against these):

• No rule-of-three. Balanced lists ("same friends, same hobbies, same her") and matched pairs ("interest or disinterest") are the classic written-by-AI tell. Break the symmetry: list two things, or four, or start a list and abandon it.
• No thesis sentences. Never summarise the insight with an abstraction ("so the loop never closes"). Land the idea through the concrete example and trust the viewer, real speakers don't label their own point.
• One self-interruption or restart per script. A "wait, no, the best example—", a trail-off ("she just sort of..."), a mid-sentence correction. Grammatically complete sentences marching in order is how essays sound, not people.
• Verbal tics go in the boring middles, not the peaks. Bolting a tic onto the rhetorical high point is where a writer puts it; speakers drop them mid-flow where nothing is happening.
• One uselessly specific detail. "A guy she dated for like five minutes in school" beats "a guy from her past". Efficiency is the biggest tell of all, one detail that does nothing for the argument is what makes it sound lived.
• Allow one near-repetition. Humans say the thing, then say it again slightly differently, because they're hearing themselves think. Keep one.
• Uneven rhythm. Cluster sentence lengths: a few stubby fragments, then one that runs on too long with "and" in it three times. Never alternate short-long on a tidy pulse.

Structure every script:

1. Hook, the first spoken line IS the hook from the idea (tighten it, keep the surprising word first). It makes a promise.
2. Deliver, pay off that promise. Build with an open loop or a countdown so they stay.
3. Payoff, resolve it. The satisfying line the hook was pointing at.
4. CTA, one line (see Step 2.4).

Never bait-and-switch: the script must deliver exactly what the hook promised.

Step 2.3: Length discipline

15–40 seconds of speech ≈ 40–110 spoken words. Count them. If it's over, cut, shorter and tighter always beats longer and padded. Aim for the low end unless the idea genuinely needs the room. If an idea's hook explicitly promises a specific runtime ("give me sixty seconds"), that's the one exception, honour the promise rather than cutting to the default range.

Step 2.4: Safety, monetisation, and the CTA

• Safety (carried from Phase 1): no medical claims, cures or "this fixes X", keep health lifestyle/habit-level. No guaranteed returns or specific financial advice, keep money to principles and mindset. Nothing that would get an account flagged or demonetised.
• Don't undercut their product: if the character sells their own thing, the script must never argue against it. Where it fits, steer naturally toward it.
• CTA: end every script with ONE short spoken CTA as the final line, if they sell a product, a soft plug that fits the character (point to the thing in their bio, not a hard sell); otherwise, a light follow/engagement line in the character's voice. Keep it to one line, and mark it so the user can delete it if they'd rather. Some hooks (a strong follow-CTA placed at the top, a pure quote clip, a joke that ends on the punchline) don't need a bolted-on closing line, use judgement rather than forcing one on every script.

Step 2.5: Output: a shootable card per video

For each script, output exactly this:

[Idea title]

• On-screen text (hook caption): "…", the muted-viewer version of the hook, short enough to read in a second.
• Setting: [carried from the idea, written so it drops into an image/scene prompt].
• Script:

  [clean spoken script here, hook first, short lines, ending with the CTA line if one's used]

• Length: ~N words (~S seconds)

Keep the script block clean and copy-pasteable, ready to drop straight into a voice tool with zero editing. No preamble, no explanation of your process, just the card(s).