The UGC
Video Playbook
Every short-form video framework we've verified, rebuilt into one plain-English manual. Part A is for filming yourself. Part B is for building UGC with AI. Same psychology — two production lines.
How to read this
Every module follows the same six beats. Once you've read one, you know the shape of all of them.
In plain English
The idea, defined without jargon.
Why it works
The mechanism — the reason it moves a number.
See it
A diagram, because some things are faster to look at than read.
Worked example
Weak version vs strong version, side by side.
Do this
The steps, in order.
Avoid this
The mistake that undoes the work.
About the confidence chips
Every framework carries its real score from our knowledge base. Read them honestly:
conf 0.90+ Verified across sources — execute as written.
conf 0.80–0.89 Strong signal — test before you scale it.
BELIEF <0.80 Contested — treat as a hypothesis to A/B, never a rule.
What's inside
What UGC actually is
The format definition most people get wrong — and why the distinction changes how you shoot.
Viewers scroll past adverts and stop for people. UGC borrows the visual grammar of a friend's post — handheld framing, direct address, imperfect lighting — so it clears the ad-filter your audience has built up. You are not fooling anyone; you are removing a barrier.
Reach without spend
Organic short-form is the one channel where a single asset can outrun a paid budget.
Trust transfer
A face and a real voice carry credibility stock footage cannot buy.
It compounds
One shoot becomes eight assets (Module 09), so cost per usable asset collapses.
Cheap iteration
Hooks swap without re-shooting the body — test the variable that matters most.
Polish signals "advert". Texture signals "person". You are optimising for the second.
Why it works — and what we won't promise
The honest version: what a system can and cannot do for you.
We do not promise views, followers, virality, client work or income. No framework can. Reach depends on the algorithm, your niche, timing, execution and luck. Anyone selling guaranteed virality is selling something we aren't. What follows is a method, not a lottery ticket.
What "working" actually looks like
- You can script in one sittingA hook and a structure chosen from a framework instead of invented from nothing.
- You publish on a cadenceProduction stops being a mood and becomes a routine you can sustain.
- You can diagnose a flopA drop-off at 3 seconds points to Module 03. A drop at 10 points to Module 07. Causes, not vibes.
- Your library compoundsEvery shoot leaves reusable assets behind instead of one disposable edit.
Filming yourself 🎥
You, a camera, and a framework. This is the highest-trust, highest-ceiling format — and the one every AI workflow in Part B is imitating. Read it even if you only plan to generate.
The Two-Hook Law
Most creators write one opening line. That's why their retention dies at second three.
Stopping the scroll and holding attention are different problems. A shocking line stops the thumb but gives no reason to stay. A slow story gives a reason to stay but never stops the thumb. You need one of each — which is why the two best hooks almost always come from different trigger categories.
The 10 triggers
Pick the emotion first, then write the line. The right-hand column tells you which job it can do.
| # | Trigger | What it does | Best as |
|---|---|---|---|
| 01 | Curiosity Gap | Opens a question the brain must close | Stop-scroll |
| 02 | Contrarian | Attacks a belief the viewer holds | Stop-scroll |
| 03 | Authority & Proof | Leads with a number or result | Either |
| 04 | Emotional | Mirrors the viewer's inner monologue | Retention |
| 05 | Listicle | Dense, scannable, save-worthy | Either |
| 06 | Question | Forces the brain to answer | Stop-scroll |
| 07 | Story | Mid-story opening they must resolve | Retention |
| 08 | Negation | Kills a belief in 2–4 words | Stop-scroll |
| 09 | Specificity | Numbers, times, dates | Stop-scroll |
| 10 | Confession | Vulnerability that earns instant attention | Retention |
“Today I want to talk about applicant tracking systems and how they work.”
One hook doing neither job. It labels the topic, so there is no gap left to close — and it opens on setup, not payoff.
“No human ever saw your resume.” → “It was scored by software first — here's the score.”
Trigger 08 (Negation) stops the thumb. Trigger 07 (Story) opens the loop. Two categories, payoff first, both under 12 words.
- Write Hook 2 firstThe loop is harder. Get the question right, then find a line that earns the first second.
- Tag both with a trigger numberIf both tags match, one of them is redundant — rewrite it.
- Cut to 12 words eachSpeakable in one breath. Lead with the payoff, never the setup.
- Use one real numberPull a digit from your actual material. Vague hype reads as noise.
Banned openers: Unlock · Discover · Dive into · Imagine · Level up · Transform · Elevate.
The labelling trap: if your hook describes what the video is about, it is a title, not a hook. Tease — never label. And never guarantee income or outcomes.
A three-part hook that borrows attention the algorithm has already validated:
- React to a viral videoOpen on something already proven to hold attention.
- Peak curiosity with a questionTurn the reaction into an open loop.
- Satisfy it with third-party authorityClose on a study, quote or statistic — not your opinion.
Watch it in practice
Hook-writing, demonstrated. Click through and watch the first three seconds specifically — that is the whole lesson.
Choosing a structure
Pick the skeleton before you write a word. Three shapes cover almost everything.
Short-form punishes hesitation. A structure removes the biggest source of it: deciding what comes next while you're writing. It also makes failure legible — if people leave at the same point every time, you know exactly which beat is broken.
The safest default and the best starting point. Name the problem in the viewer's own words, show the cost of ignoring it, deliver one mechanism, then close.
Promise a number in the hook and deliver it fast. Cut on every number. Strongest item first — people leave before the end, so never save your best for a payoff they won't reach.
Show the finished result first, then walk back through the steps. Withholding the payoff is the most common tutorial mistake — viewers stay for a result they've already seen.
Record the full topic once as long-form, then cut only the hook section as your reel and end on a CTA to the full video. One recording feeds both surfaces.
A 45-second video that explains ATS history, then definitions, then finally the tip at 0:38.
Tutorial shape with the payoff buried. Most viewers never reach the only useful part.
A 22-second Problem–Solution: “No human saw it” → what that costs you → the one fix → where to do it.
One mechanism, front-loaded. Every second is doing a job the structure assigned it.
The retention build
The most aggressive structure we have — engineered for vertical short-form.
- Cold open on a micro-expressionStart mid-reaction. No intro, no greeting, no setup.
- Stack proof roughly every 1.4sEvidence arriving constantly is the retention mechanism.
- Explain the mechanism visuallyA diagram beats a description. Draw it if you have to.
- Montage the modern applicationFast cuts proving it still applies now.
- Drop to a whisper to closeFalling energy at the end reads as sincerity.
- End on a curiosity-gap CTATease a future part instead of asking for a follow.
This is the single most mechanical lever in the playbook. It needs no better writing, no better camera and no better idea — only better cutting. If a take runs long, the fix is a B-roll insert, not a re-shoot.
One unbroken 9-second shot explaining three points.
Four times over the ceiling. Nothing new arrives on screen for nine seconds.
Same 9 seconds cut into: 2.0s face → 1.6s B-roll → 1.8s face → 1.9s screen-recording → 1.7s face.
Identical script, identical footage. Only the cutting changed — and the density with it.
Leaked object · Expression · Angle · Keyword · Status reversal — the editing grammar for news, culture and reaction content.
The job-search framework
Derived from top performers in our own category. Start here for career content.
- Hook a cold audienceAssume zero context and zero goodwill.
- Name the common enemyShared frustration bonds faster than shared aspiration.
- Deliver ONE concrete tacticOne. Specific. Usable today. Not a list of principles.
- Weave the product in naturallyIt appears as how you did the thing — never as an ad break.
- Establish credibilityA number, a result, a receipt. Real ones only.
- Drive engagementAsk a question the comments can actually answer.
“…and that's why you should sign up for our resume tool — link in bio, it's only $7.”
A hard pivot into an ad. The viewer feels the switch and leaves at the seam.
“I pasted the posting in, saw which terms were missing, and fixed those three lines.”
The product is the method, not the pitch. Nothing to resist because nothing is being sold at them.
Career content reaches people under real financial stress. Never guarantee employment. Never invent rejection statistics. Never imply a paid tool removes the need to be qualified. Say “a real person did X”, never “you will earn $X”.
Editing for retention
Where the script stops mattering and the cut takes over.
Our master editing framework: data-driven editing, scripting and retention rules synthesised from top-creator analytics. It governs the edit, where the other frameworks govern the script.
The retention floor — non-negotiables
Audio first
Bad audio kills retention instantly. It outranks camera, lighting and editing. Get the mic close.
Always caption
A large share of viewers watch muted. No text overlay means an invisible video.
Cut ruthlessly
If it can be shorter, make it shorter. Length is not value.
No slow build
Don't build up to the point. Open on it.
Some data suggests a CTA inside the first 5 seconds improves retention and conversion. This directly contradicts the Seamless Loop in Module 08 (conf 0.97). We are not going to pretend that conflict doesn't exist: higher confidence wins by default, so loop first and run early-CTA as the challenger. This is the lowest-confidence item in the playbook.
A two-pass edit before publishing: (1) delete every sentence that isn't load-bearing, (2) reduce reading complexity. Brevity and a low reading grade beat cleverness on short-form.
P.A.S.T. — VERVE's short-form spine
Pain
Name the thing they already feel. Their words, not yours.
Agitate
The cost of leaving it unsolved. Brief — this is the shortest beat.
Solution
One mechanism. Shown, not described.
Trap Door
Don't land — drop them into the next loop (see the Seamless Loop, Module 08).
Scope note, stated honestly: VERVE's cut cadences and micro-expression prescriptions come from our uploaded VERVE document — they are that document's claims, not independently verified platform laws. Do not force a 1.4s cut rhythm onto a quiet, emotional or long-form piece. OmniCut is the operating system; VERVE is a retention module you switch on when the content suits it.
The seamless loop
Replace the ending with a beginning.
- Write the hook firstThe last line has to land on it, so it has to exist first.
- End on the hook's setupYour final sentence should make sentence one feel like the next one.
- Cut the goodbyeNo sign-off, no “follow for more” — both announce the ending.
- Match the framingClose on a shot near the opening frame so the cut is invisible.
“…so that's how the ATS works. Follow for more job tips!”
A hard stop. The sign-off tells the viewer they're allowed to leave — so they do.
“…which is why no human ever saw it.” (→ loops into: “No human ever saw your resume.”)
The last line lands exactly on the hook. The replay is seamless and the watch time compounds.
Don't loop a video whose only job is conversion. If you need a click, ask for it — a loop keeps people watching, not clicking. Loop for reach, CTA for action.
One shoot, eight assets
Where your cost per usable asset collapses.
| # | Asset | What it's for |
|---|---|---|
| 1 | Master short | The primary publish. |
| 2 | Hook cut | Alternate opener on the same body — the cheapest A/B you can run. |
| 3 | Proof cut | Leads with the receipt instead of the claim. |
| 4 | Framework cut | The teaching moment — the one people save and share. |
| 5 | Objection cut | Answers the top comment objection head-on. |
| 6 | CTA cut | Conversion-weighted variant for paid placement. |
| 7 | Long cut | Feeds YouTube / LinkedIn from the same session. |
| 8 | Cover frame | The still. Governed by our thumbnail framework, not this playbook. |
Production floor
The six mistakes that kill videos, and how to study without copying.
| The mistake | The fix |
|---|---|
| Slow hook | Open on the point. Delete your first sentence — it's almost always throat-clearing. |
| No text overlay | Caption everything. Assume muted. |
| Poor audio | Get the mic close. This outranks every camera upgrade you're considering. |
| Too long | If it can be shorter, make it shorter. |
| No CTA | Tell viewers what to do — or loop deliberately (Module 08). Choose one. |
| Ignoring comments | Engagement in the first hour matters. Be there for it. |
Study the right inputs before you create. Filter by: niche targeting · engagement thresholds · creator-size limits · recency window · topic exclusions · priority signals. Studying the wrong reels is why most “inspiration” scrolling produces nothing.
Consistent talking-head reels + efficient scripting + conviction on camera — a posture that reduces reliance on cold outreach and paid ads. Mid confidence: test it, don't bet the business on it.
Study frameworks, never copy content. Analyse structure, pacing and hook shape. Never repost another creator's footage, branding or copy — and never put an identifiable person in a paid ad without their consent. A watermark is not a licence.
Building with AI 🤖
Same hook laws, different production line. Everything here is built from tooling we have actually tested, with the real costs we measured — not vendor marketing.
When to use AI — and when not to
The decision is about trust, not budget.
AI is right
B-roll you can't shoot · scenario footage (a courier at sunrise, a lineworker on a pole) · hook A/B at volume · scroll-stopping stills.
AI is wrong
Anything needing earned trust · your personal story · on-screen text · a specific real product UI · any claim that needs a receipt.
Generated footage carries no lived authority. The moment a viewer suspects the person on screen doesn't exist, every claim attached to them inherits that doubt. So use AI where the footage is illustrative, and use your own face where the footage is evidentiary.
The model stack & real costs
Measured on our own account — not copied from a pricing page.
| Model | Best for | Measured cost | Notes |
|---|---|---|---|
| gpt_image_2 | Photoreal UGC stills | 1 credit @1k/med · 2 @2k/med | Works on the free tier. The workhorse. |
| gpt_image_2.5 | Highest-fidelity stills | 3 credits @2k/high | 4k, transparent bg, Flare/Sunburst. Paid plan required. |
| kling3_0_turbo | Cheap image→video | 7.5 credits / 5s 720p | Budget motion test. |
| marketing_studio_video | TikTok/Reels ad video | 60 credits / 12s 720p | Hooks, settings, and reference-driven ad rebuilds. |
| virality_predictor | Pre-spend QA | Scores an existing video | Hook strength and retention risk before you pay for distribution. |
Two traps we hit ourselves:
1. Quality tier is a cost multiplier. A 30-image run at 2k instead of 1k doubles the bill. Always run a cost preflight before a batch.
2. Free tiers cap concurrent submissions. Batch items get silently rejected — not failed loudly. Check what actually came back and retry the gaps.
Prices and tiers move without notice. Treat this table as a snapshot dated on the cover, and re-check before any large spend. We would rather show you a dated number we actually measured than a current number we guessed.
The Base + Composite method
The one split that separates usable AI creative from obviously-AI creative.
- Write a scene brief, not an adDescribe the subject, their reaction, and a proof object in focus. Never ask for words.
- Reserve negative spaceInstruct the model to keep one side clean and uncluttered — that's where type will land.
- Force the palette in-sceneAsk for your accent colour as a real object — a mug, a hard hat, a bollard. It reads as photography, not a filter.
- Composite typography locallyYour display font, your exact hex, your handle placement.
“A medical courier with text saying $65 AN HOUR in bold yellow letters.”
Asking for text guarantees mangled letterforms, and the colour arrives as a filter over everything.
“A courier in navy polo and nitrile gloves loading a transport cooler; biohazard bag and clipboard in focus; keep the left 55% clean and dark; one saturated yellow object in scene.”
Photograph only, space reserved, palette present as a real object. Type gets composited after, perfectly.
Always block these: any text, letters, numbers, words, captions, watermarks, logos or signage · malformed hands · fake stock grin · collage or split-screen · borders. Most “obviously AI” giveaways come from exactly these.
Script → AI video
AI does not exempt you from Part A.
- Start at the Two-Hook LawModule 03. The hook discipline is identical — only the camera changed.
- Pick a structureProblem–Solution is the safest AI skeleton: it needs the fewest believable human beats.
- Storyboard to the 2.1s ceilingPlan cut points before generating. Generate only shots you'll actually use.
- Generate bases in batchPreflight the cost, then batch. Expect concurrency rejections on free tiers and retry the gaps.
- Composite type and captionsYour fonts. Always captioned — muted viewing applies to AI video too.
- Run the QA gateModule 15. Score it before you spend distribution on it.
It moves the hardest variable — structure — from guesswork to copying something that already worked. Note the sharp edge: avatars or products linked to a reference are not applied automatically. Pass them explicitly or you'll get the structure with the wrong subject.
The QA gate
Generation is cheap. Distribution is not.
| Check | It passes when… |
|---|---|
| Three-second test | A stranger who saw 3 seconds can tell you what it's about. |
| Muted test | The meaning survives with the sound off. |
| 1-inch test | The cover still reads at thumbnail size. |
| Hands & faces | No malformed hands, no uncanny expression, no melted background text. |
| Claim audit | Every number on screen traces to a real, citable source. |
| Rights audit | No third-party faces, logos or footage you don't have rights to. |
| Virality pre-score | Hook strength and retention risk checked before you pay for reach. |
A failure from our own library: clips saved as “branded” turned out to be other creators' footage with a watermark added. Fine to study. Not fine in a paid ad. Audit where every asset actually came from before it enters a campaign.
Disclosure & compliance
The part that protects you.
Label AI content
Most major platforms require disclosure of realistic AI-generated media. Use the platform's own toggle — a caption is not a substitute.
Respect likeness
Never generate a real, identifiable person. Never put a stranger's face in a paid ad. Consent is required, not implied.
Logos and brands
Use only official logos you have the right to use. Never invent a logo. Never imply a partnership that doesn't exist.
Claims discipline
Real numbers only, attributed. No income guarantees, no fabricated statistics, no invented testimonials.
Disclosure requirements are changing faster than this document can. Check each platform's current policy before a campaign — that check is your responsibility, not this playbook's.
The compliance floor isn't a legal formality. It's the difference between a channel that compounds and one that gets removed.
OmniCut — the source-to-edit operating system
Not a script writer. The layer that turns real source material into an executable edit.
Most AI video systems run: prompt → generate something visually interesting. OmniCut inverts it. It refuses to ask “what cool visuals can we make?” until it can answer “what is this video actually trying to accomplish?” — the hook, promise, audience pain, offer, proof, tension, payoff, CTA and pacing. Visual decisions are downstream of that, always.
“Make a cinematic video about AI infrastructure. Data centres, glowing servers, drone shots.”
Starts at GENERATE. Produces attractive footage that argues nothing, because no one established what the viewer should believe by the end.
“Thesis: power changed form. Beat A kills ‘AI = app’. Beat B installs ‘AI = infrastructure’. Beat C reveals the consequence. Visual progression: interface → interface disappears → substations → server racks → ownership map.”
Starts at UNDERSTAND. Every shot now has a job in an argument, so the B-roll becomes reasoning instead of decoration.
The three production modes
Mode is chosen first, because it changes the entire editing logic — not just a few prompts.
| Mode | Primary job | Creative priority |
|---|---|---|
| AI Influencer | An AI person / avatar leads the video | Human believability — personality, micro-expressions, delivery |
| Faceless YouTube | Documentary / explainer / storytelling | Story, B-roll density, information design, retention |
| AI Video Editing | Transform or reconstruct an existing video | Precise timestamp and frame-level execution |
Woman says: “Nobody tells you this about AI.”
That is dialogue generation. The model invents the performance, so identity and delivery drift between clips.
28-year-old creator, minimal black desk, medium close-up, 35mm equivalent. Begins looking at the laptop, not the lens. On “Nobody,” glances directly into lens. Brows rise slightly; lips stay neutral, no smile. Small forward lean on “AI.” Camera pushes in 6% over the final 0.8s.
That is performance direction. The person is treated as part of the editing system, not an output of it.
Source ingestion comes before creativity
OmniCut first builds a source-of-truth package: video, transcript, screenshots, website, product/offer details, audience, length, platform, CTA, brand tone, reference examples. Missing information never blocks production — it gets labelled.
Confirmed
Present in the source. Usable as fact.
Unknown
Absent. Stays absent — never quietly filled in.
Assumption
A reasonable creative inference, explicitly marked as one.
Inferred opportunity
“Insert verified customer-result screenshot here if available.”
Transcript says “our customers are getting results” → edit instruction: “Show three customer testimonials.”
No testimonial assets were supplied. The instruction invents evidence and an editor will either fake it or stall.
Transcript says “our customers are getting results” → Inferred editing opportunity: insert a verified customer-result screenshot here if available.
The gap stays visible. Nothing is fabricated and the human knows exactly what to supply.
The website intelligence layer
If a site or landing page is supplied, OmniCut doesn't summarise it — it converts it into video-relevant intelligence: offer, visible pricing, target customer, benefits, features, proof points, testimonials, claims, differentiators, CTA language, brand tone, funnel structure, visual identity, product hierarchy.
The video tells OmniCut what is being said. The website tells it what the business needs the video to accomplish.
The truth / verification layer
Five categories that must never silently merge: confirmed source evidence · user claims · external research · creative inference · generated dramatisation. You may dramatise a concept. You may not make a fabricated document, headline, statistic or quotation look like evidence.
How the other frameworks stack on top
The operating system. Source → understanding → direction → execution.
Short-form retention layer (P.A.S.T.). Switch on when the content suits it — Module 07.
The packaging layer. Sits outside the timeline, derived from it.
The conversion layer — “messaging is the targeting.”
Same operating system, different objective. A documentary and a direct-response ad run the same pipeline with different modules engaged.
OmniCut — transcript-first direction
The spoken message controls the visual system. Never the other way around.
“AI is not an app. AI is becoming infrastructure. And whoever owns the infrastructure owns the age.” → three shots of AI imagery.
Treated as sentences. The visuals repeat the same idea three times and the argument never visibly moves.
Beat A destroy old belief (AI ≠ app) → Beat B install new belief (AI = infrastructure) → Beat C reveal consequence (ownership = power).
Visual: chatbot interface → interface disappears → data-centre aerial → substations → server racks → ownership map.
Treated as beats. The picture carries the logic, so the viewer understands before they finish listening.
Frame-aware analysis
Once meaning is mapped, OmniCut asks of every moment: what should the viewer be seeing, reading, hearing and feeling?
| Segment | What OmniCut evaluates |
|---|---|
| 0:00–0:03 | First-frame strength · hook clarity · curiosity · visual comprehension · scroll-stopping potential. The first frame has a job before the sentence is even processed. |
| 0:03–0:10 | Why shouldn't they leave? What promise is established? What can be compressed or cut? Where do captions get more aggressive? Where does proof begin? |
| Core | Every teaching / story / selling / proof beat gets its own visual language: talking head, archival, B-roll, screen recording, diagram, chart, screenshot, reconstruction, motion graphic, typography. |
| Payoff | The viewer receives what the hook promised. This beat gets extra visual and sonic weight. |
| CTA | Treated as an edited sequence, not a sentence — placement, wording, visual treatment, music and camera behaviour, and whether it's direct, soft, comment-driven or community-driven. |
Hook Surgery
- Best overallThe strongest balance of reach and accuracy.
- SaferLower ceiling, lower risk — for sensitive or regulated topics.
- Aggressive / viralMaximum stop-power. Check it against the claim audit.
- CuriosityPure open loop, minimal information.
- Direct-responseBuilt to convert rather than to reach.
Each version must carry the spoken line, on-screen text, first visual, sound cue, and the reasoning behind the construction.
“Most people are using AI wrong.”
A line, not a hook. No first frame, no typography, no sound design — the edit is left to guesswork.
Voice: “Everyone thinks this is the AI revolution.”
Frame 1: chatbot interface filling screen.
0.35s: red X slams over it.
Voice: “It isn't.”
0.8s: smash cut to a vast night data centre.
Text: THIS IS.
Sound: UI click → bass hit → low industrial hum.
The hook now exists in language, imagery, typography and sound at once. Nothing is left to interpretation.
The beat-by-beat editing matrix
The operational heart of the system. Every row must be executable without follow-up questions.
| Timecode | Transcript / VO | Visual direction | Edit instruction | On-screen text | B-roll / asset | Sound | AI prompt |
|---|---|---|---|---|---|---|---|
| 00:00–00:02 | spoken phrase | what the viewer sees | exact edit behaviour | caption | asset required | audio treatment | generation instruction |
Where exact timecodes exist they are used; otherwise the system falls back to Beat 1 → Beat 2 → Beat 3. The table is not the point — execution density is.
The specificity rule
| ✗ Weak | ✓ OmniCut |
|---|---|
| Add B-roll. | Add a 1.5s close-up screen recording of the pricing section, push 100% → 118%, underline the key claim in yellow, cut on the bass hit. |
| Add captions. | Bold white kinetic captions, 4–6 words per line; red emphasis on pain words, green on outcome words. |
| Make the CTA stronger. | Duck music 40%, punch camera in 12%, hold CTA centre-screen 1.2s, then trigger a DM-notification visual. |
Specificity × timing × source accuracy × visual purpose = execution quality.
OmniCut — continuity, sound & the deliverable
What keeps ten generated clips feeling like one film — and what you actually hand over.
Without continuity locking, segmented AI generation degrades fast: clip 1 brown jacket, clip 2 black jacket, clip 3 a different face, clip 4 sunset becomes midday, clip 5 cinematic becomes handheld. OmniCut treats those as continuity failures, not acceptable model creativity.
- Lock the bible before generatingIdentity, wardrobe, light and lens decided once, written down.
- Chain the scene statesPrevious scene state → current scene → ending state → next scene state.
- Separate timeline from generationAn editor needs “cut here.” A model needs “generate this exact shot.” Write both.
- QA against the bibleIdentity, wardrobe, environment, camera, typography, audio — before assembly, not after.
Captions are an editing layer
Not transcription decoration. Style, font direction, emphasis, highlighted words, punctuation and timing are all decisions — and typography should carry the argument's hierarchy.
AI IS NOT AN APP
↓
IT'S INFRASTRUCTURE
↓
POWER → COMPUTE → INTELLIGENCEB-roll is evidence, not wallpaper
Each asset is additionally marked as required, optional, AI-generated, or user-must-supply — so nobody discovers a missing asset mid-edit.
Sound is its own storytelling layer
Not “add some cinematic music.” The layer defines music style, SFX, whooshes, clicks, risers, bass hits, emotional audio moments, CTA treatment — and silence as a tool.
Music swells continuously under the revelation, plus an impact hit on the key line.
Everything is loud, so nothing is emphasised. The important sentence competes with the score.
Music builds → narration nears the revelation → music disappears → the sentence lands dry → one second of silence → visual reveal → score returns.
The silence gives the line more authority than any impact sound could.
Platform optimisation comes last
After the edit concept, not before it: aspect ratio, duration, opening speed, caption density, CTA style, safe zones, cover-frame selection, retention risks, likely drop-off points. You don't make one master and crop it everywhere — a YouTube documentary can breathe; a Reel cannot use the same opening architecture.
OmniCut separates the message from the platform's expression of the message.
The mandatory 12-part deliverable
Every complete OmniCut analysis resolves into the same package — and it ends with something another system can execute, not just read.
| # | Deliverable |
|---|---|
| 01 | Production mode selected |
| 02 | Source material summary |
| 03 | Big creative direction |
| 04 | Hook surgery |
| 05 | Frame-by-frame / beat-by-beat editing script |
| 06 | AI editing master prompt |
| 07 | Scene generation prompts |
| 08 | Caption system |
| 09 | B-roll & visual asset list |
| 10 | Sound design |
| 11 | Platform optimisation |
| 12 | Final copy/paste AI editing prompt |
The final master prompt
Thousands of words of reasoning compress into machine instructions covering style, pacing, captions, colour, camera, transitions, B-roll behaviour, sound, platform formatting, CTA handling and export.
Preserve source dialogue exactly. Remove dead air. Maintain subject identity across all generated segments. Use documentary-grade B-roll only where it advances the argument. Keep archival material visually distinct from generated material. Match every conceptual shift with a visual shift. Use restrained kinetic typography for thesis statements only. Do NOT fabricate documents, headlines, statistics or quotations. Maintain the Continuity Bible across all generated segments. Export: [MASTER + PLATFORM VERSIONS + CAPTIONS + PROMPTS + PROJECT DATA]
The core unit is the production beat
A normal script says what to say. A storyboard says roughly what to show. An editing brief says roughly how it should feel. A generation prompt says what image to make. OmniCut combines all four — and if your output only does one of them, you have written a script, not an OmniCut blueprint.
See AI video workflows running
I Automated UGC Content with Sora 2 AND Veo 3.1 in n8n▶ Nate Herk | AI Automation
I Built the Ultimate UGC Content System with AI Agents▶ Nate Herk | AI Automation
This AI Makes My UGC Ads Every Day▶ Website Learners
Clone Any Viral Reel With Your Face▶ Website Learners
Create Viral Reels in Seconds — No Script, No Edits▶ Website Learners
Build a YouTube Thumbnail Generator with n8n▶ Website LearnersGlossary
The opening. Two jobs: stop the scroll (second 0), then open a loop (seconds 1–3).
An unanswered question that makes leaving feel like missing something.
How often something new arrives on screen. Attention decays with sameness, not time.
Any shot that isn't the talking head. Its main job is resetting attention at a cut.
An ending written to set up the opening, so the replay is invisible.
The eight separate deliverables cut from one recording session.
Generate the photograph with AI; add all typography yourself.
The physical thing in frame that makes a claim credible — a cooler, a manifest, a screen.
The list of things the model must not produce. Text is always on it.
The real cross-source score from our knowledge base, shown so you can weigh the advice.
Resources
JobHacki tools
Any video → a timestamped, cut-by-cut edit blueprint.
✍️OmniScriptINCLUDEDTurn any idea into a viral-ready script with hooks.
📄Resume BuilderSCORING FREEPaste the posting, get scored against it.
💼Job DirectoryFREELive roles across 12+ ATS systems.
📚Playbook Directory$7 ALL-ACCESS100+ step-by-step income playbooks.
🎯Readiness SimulatorFREEPaste a job link — get tested + a study path.
Where you'll publish
Raw, authentic compositing wins here.
Instagram ReelsHeavier produced poster aesthetic wins here.
YouTube ShortsFeeds the long-form extraction loop (Module 04).
LinkedInUnderrated for career-niche short-form.
Models referenced in Part B
The photoreal base generator in Module 13.
Anthropic — ClaudeCREDITSScripting and structure work.
Where this came from
25 video frameworks from the JobHacki knowledge base, deduplicated and synthesised: hook rules, structure libraries, editing engines, retention research, discovery filters and the measured model stack. Cover frames are governed by our thumbnail framework, not this playbook.
Changelog
| Version | Date | Change |
|---|---|---|
| v4.1 | September 20, 2026 | Published free and public. Canonical URL set to the real static path, schema marked accessible-for-free, funnel re-pointed to free access with All-Access as the upsell. |
| v4.0 | September 20, 2026 | Real JobHacki logo. Replaced the placeholder OmniCut recipes with the canonical OmniCut framework (Modules 17-19): 9-stage pipeline, 3 production modes, source-truth labelling, hook surgery, beat matrix, specificity rule, Continuity Bible, sound layer and the 12-part deliverable. VERVE documented as a retention module with P.A.S.T. |
| v3.0 | September 20, 2026 | Added the Arsenal CTA, tool link-cards with official logos, clickable video references, Module 17 (scripts / prompts / OmniCut recipes), original cover art, and full SEO markup. |
| v2.0 | September 20, 2026 | Full redesign. Four-role type system, explanatory diagrams, plain-English definitions, worked examples and a glossary. Six-beat module rhythm. |
| v1.0 | September 20, 2026 | Initial release. 25 frameworks synthesised. Part A + Part B. Model costs measured on-account. |
What triggers a revision
A new verified framework · a confidence score moving · model pricing or tier changes · a platform changing disclosure or format rules · a framework being contradicted by new data — in which case we mark it, we don't quietly delete it.
This playbook is free. So are 100+ more.
No paywall on this one. Create a free account and the whole Playbook Directory opens up — plus the tools below.
Land your dream job. Start your dream business.
Any video → a timestamped, cut-by-cut edit blueprint.
Turn any idea into a viral-ready script with hooks.
Recruiter-tested one-page resume, auto-built from your LinkedIn.
Paste a job link — get tested + your fastest study path.
252 vetted tools, prompts, repos and GPTs across 86 categories.
338,947+ live ATS jobs across 18,000+ company boards.
100+ grounded, step-by-step income playbooks.
Every future revision of this playbook included.
Join free today — All-Access is $7 for 7 days, then $27/month. Cancel anytime.