He Rambled Into His Phone. It Came Back as a Real Video.
Talking-head run sheet · NextLevel Digital · working title, not locked
10–13 min
Sean talks to camera — no screen share
Visuals cut in later
Cut hard after the payoff
🖥️ what's on screen
🎙️ pull any of these lines
★ the one line that has to land
➡️ roll into the next beat
⚠️ careful here
💬 the soft ask
Cold open — play the memo, then cut
0:00 – 0:15 · no intro, no hello
🖥️
Amir's real voice memo playing over a waveform — about 8 seconds of "…using AI outbound dialers is very very dangerous…" Then a hard cut to the finished Compliance video, same idea, narrated and animated.
🎙️
- ★ "That's my business partner talking into his phone. This is what came back." say it flat and let the cut do the work — the contrast IS the hook
- "Two minutes and twenty seconds of rambling. No camera, no editor, no video software."
- "He wasn't reading anything. He hit record and talked."
- (if you want the number version) "Two twenty of him. One forty-three of that."
➡️
"Same words. Same guy. One of them took him two minutes — the other one took a machine."
What you're getting, and why listen to me
0:15 – 0:50
🖥️
Deck: "No camera. No editor. No video software."
🎙️
- "I'm going to show you every piece of how that gets built. The look, the voice, the timing, the render."
- ★ "I build the CRM and the marketing systems that insurance agents and agencies run their lead flow on. We ship product constantly — and every time we ship, somebody has to make training videos for it. That's the problem this solved." the credibility line — say what you build, never what you're not, and no client counts on camera
- ★ "And before you tune out because this sounds technical — you're not writing any of this. Claude Code writes all of it. You're describing what you want, in English, and looking at what comes back." this is the line that keeps the non-developer audience past minute five — the SEO brings in people who are not coders
- "This isn't a magic button. It's a machine we built once, and now it prints these."
- (no mention of the free PDF here — the only CTA in this video is the last thing you say)
⚠️
No client counts on camera — "agents and agencies," never a number. And never phrase credibility as what you aren't.
➡️
"But the part that surprised me wasn't the animation…"
Plant the loop RE-HOOK
~0:50 · one sentence, then move on
🎙️
- ★ "…it's that the animation knows exactly when every word is spoken. Down to the millisecond. And it figured that out by listening to itself." this is the open loop the whole video pays off — don't explain it yet
- "I'll come back to that, because it's the piece that makes this a machine instead of a one-off."
➡️
"First — what he actually handed me."
Piece one — the raw material
the input
🖥️
Deck: the real transcript on the left with the ugly bits struck through, the cleaned script on the right.
🎙️
- "No script. No prep. Two minutes on compliance because that's what was on his mind."
- "Look at the actual transcript. 'Very very dangerous.' 'Compliancy rules.' 'That's all I got for you.' That's not a writer. That's a guy in his office."
- "Grammar got tightened. The sentence that doubles back on itself got untangled. The sign-off got cut. That's it."
- "We also have house terms — words we always use and words we never use. Those go in as a rule, and the machine holds them every single time, across every video, without anyone remembering to check."
- (don't say the banned term itself on camera or put it on a slide — the rule is the point, not the word)
- ★ "It didn't rewrite him. If it had, it'd sound like a robot wrote it. It sounds like Amir, cleaned up." this is what separates it from AI slop — say it out loud
➡️
"So now we've got words. Next piece is the one everybody skips — and it's the reason this doesn't look like every other AI video on the internet."
Piece two — the brand kit
why it looks like your company
🖥️
Deck: the five brand colors, the three fonts. Then flip to the real stylesheet file, then a finished frame next to the actual app UI.
🎙️
- "This is the difference between something that looks like your company and something that looks like a template with your logo pasted on it."
- "It's one file. Colors, fonts, the shape of a card, the glow on a badge — written down once."
- "Ours is the real thing. The navy off our website, our blue, the actual logo, real screenshots of the actual app."
- ★ "You build this file one time. Every video after that inherits it for free." the entire leverage argument in one sentence
- (honest) "This part took real back-and-forth. It was not one prompt."
- (try it) "Hand it your hex codes, your two fonts, your logo — ask for one stylesheet you'll reuse. Then you never explain your brand again."
➡️
"Now it looks right. It still can't talk."
Piece three — the voice
and the trick nobody mentions
🖥️
Deck: the two script files side by side. Zoom the one word that differs.
🎙️
- "The narration is ElevenLabs — a voice called Jessica. Same voice across all three videos, so they sound like one product instead of three."
- "Here's the trick. Every script exists twice. One copy for humans, one for the robot."
- "In our world the word is 'lead' — a sales lead. Text-to-speech reads it 'led,' like the metal. Every single time." (pause, let it land)
- "So the robot's copy spells it l-e-e-d. Now it says it right — and the captions on screen still show the correct word, because they're built from the human copy."
- ★ "Two files, one word different. That's the gap between 'pretty good' and 'I can actually ship this.'" the specific, memorable detail people screenshot
- (try it) "Respell anything the voice fumbles — brand names, industry terms, acronyms. Caption from the clean file."
➡️
"Alright. Words, look, voice. Now the part I promised you."
Piece four — the timing THE TWIST
slow down here · this is the video's best idea
🖥️
Deck: the word strip with timestamps, then the real scene file showing a slide anchored to a phrase, then the finished video hitting that exact beat.
🎙️
- "Normally animation means a timeline. Drag a thing, nudge it, watch it, nudge it again. That's the part that eats your whole day."
- "We never touch a timeline. Once the voice is generated, we run it back through a transcriber — it listens to its own voiceover — and gets the exact moment every word is spoken."
- "Then slides aren't pinned to a clock. They're pinned to words. This one starts when she says 'the most important rule.' That one starts on 'eight or nine.'"
- ★ "So if you change the script, you don't re-time anything. You re-run it and the whole video re-times itself." the payoff of the loop you opened at 0:50 — this is the moment
- (honest) "It's fussy. The transcriber writes numbers as digits — so if a slide is waiting on the words 'twenty to thirty,' it never fires. You check that before you render."
➡️
"So now there's a web page that knows how to play itself. Problem is, it's still a web page."
Piece five — the render
web page → mp4
🖥️
Deck: the five-step chain — page, seek, shot, mix, mp4. Then the render actually running, frames ticking.
🎙️
- "The video is a web page. To turn it into a file, we open it in a browser with no window, jump it to a timestamp, screenshot, jump, screenshot — a few thousand times."
- "Then glue the frames to the voice, slide a music bed underneath at about five percent, whoosh on the cuts, and out comes an mp4."
- ★ "And this is why it can't go out of sync. It's not recording a screen — it's computing what the screen looks like at each moment. Slow computer, fast computer, doesn't matter. The picture literally cannot drift off the audio." everyone who's screen-recorded a laggy demo feels this in their bones
- (try it) "Headless browser screenshots into ffmpeg. Least glamorous part, most reliable part."
➡️
"Which brings me to the thing that actually matters — and it's none of the five pieces I just showed you."
The payoff — it was never about one video
the reframe
🖥️
Deck: all three finished videos side by side with their real runtimes.
🎙️
- "Building the first one was real work. Hours of it."
- "The second one was a new script into the same machine. So was the third."
- "Three finished videos in one working session — and most of that session was the first video."
- ★ "The thing you build isn't the video. It's the thing that makes the next video." the whole philosophy of the channel in one line — land it and pause
- "And when Amir wants a line changed, nobody re-records and nobody re-edits. You change a text file and run it again."
- (the product tie-back — one sentence, no pitch) "This is how the training videos get made for the agencies running our CRM. Every time we ship something, this is what makes the video nobody wants to make."
➡️
"Now — I'd be lying if I said this replaces everything."
Where it doesn't work
say these out loud — they buy the trust back
🖥️
Deck: the three limits.
🎙️
- "It's not a talking head. If somebody needs to trust a person, put a person on camera."
- "It doesn't replace a real screen recording. Sometimes people need to watch the actual clicks."
- "And the first build is genuinely work. The leverage shows up on video two, not video one."
- ★ "What it's great at is the explainer in the middle — the one nobody ever gets around to making." names the exact job to be done, so viewers self-identify
➡️
"So here's what I'll do for you."
Soft ask, then cut
two lines maximum
💬
"Everything I just showed you — the prompts, the folder structure, the tools, the order to do it in — is a free PDF in the description. No email. Go take it."
optional second line: "If you want to see what we build with this stuff, the rest of the channel's right here."
⚠️
Stop recording immediately after this. No recap, no goodbye, no subscribe speech.
Screen capture list — for Jazmin's edit
Sean is talking to camera, not screen-sharing. Everything below has to be captured separately and cut in. The 🖥️ line on each beat says what should be on screen at that moment; this is the shot list to go get.
| Beat | Capture | Status |
|---|
| Cold open | The raw .m4a playing over a waveform — first ~8 seconds only | File exists, needs the screen treatment |
| Cold open | The finished Compliance MP4, same moment | ✅ on the Desktop |
| Claude Code itself | A real terminal session running one of the prompts, with the real response coming back. The title says Claude — the tool has to appear on screen. This is the one genuinely missing shot. | ⚠️ Not captured |
| The raw material | The actual transcript file open — not the deck's version of it | ⚠️ Not captured |
| The brand kit | The real stylesheet scrolling | ⚠️ Not captured |
| The voice | The two real script files side by side | ⚠️ Not captured |
| The timing | The word-timing file, and the scene file showing a phrase anchor | ⚠️ Not captured |
| The render | The render actually running, frames ticking | ⚠️ Not captured |
| Payoff | All three finished videos playing | ✅ on the Desktop |
The deck is the designed layer — headlines, diagrams, the before/after. The real screens are the proof layer. Both go in; the deck alone is a recreation, and recreations are the thing we don't ship.
Before you hit record
- Get Amir's OK — and be specific about what you're asking. This puts his unpolished voice memo on YouTube, his first name on screen, his speech errors struck through in red, and the word "rambling" in the title. Any one of those is fine with his blessing and awkward without it. Show him the deck before you shoot.
- Air only the first ~8 seconds of the memo. Compliance optics: later in that recording he uses a lead-type term we don't use and describes dialer practices in blunt language. On a NextLevel-branded channel that's a frame a competitor or a compliance officer can screenshot. The opening seconds make the whole point. Clip there and stop.
- No client counts on camera. "Agents and agencies," never a number.
- Don't name the underlying vendor. You scrubbed it from the product narration on purpose — don't undo that on camera while the app is on screen.
- No cost claims and no invented counterfactual. Don't say what an editor "would have cost." The honest contrast is that these videos wouldn't have gotten made at all.
- Files you'll need open: the three raw
.m4a memos in Downloads, the three finished MP4s on the Desktop, the project folder, and this deck.
Titles — for Jazmin's read before locking
| Option | Targets | US demand |
|---|
| He Rambled Into His Phone. Claude Made It a Real Video. | how to create ai videos | 25K/mo · comp 27.9 |
| How to Make AI Videos That Actually Look Like Your Brand | how to create ai videos | 25K/mo · comp 27.9 |
| I Turned a 2-Minute Voice Memo Into a Finished Video (No Editor) | ai storytelling | 48K/mo · comp 26.5 |
Thumbnail direction for Jazmin (she owns it): the tension is ugly input → polished output. A phone-recording waveform against a frame of the finished branded video. Text shouldn't repeat the title.
Don't chase "claude code tutorial" — 115K US searches but the results are Anthropic's own channel and creators with 200K–3M subscribers.