SHORT-FORM VIDEO PRODUCTION STUDIO
We find the ideas worth sharing in your long-form videos, piece together the moments that matter, and turn them into polished, ready-to-post Shorts and Reels. No editing on your side.
BEFORE STEP 0
−1 · The read
The whole episode, read end to end. Twelve places where a thought actually starts and finishes. One picked — and it wasn't sitting in one piece.
THE IDEA, NOT THE TAKE
Not a cut — an assembly
The promise at the top of this page — "I go through all of it and pull out the moments that actually hold attention" — is what this clip is built to prove, not just say. Three spoken moments, none of them adjacent, are spliced into one continuous idea: 74 seconds of talk sits between the first two pieces, 20 seconds between the last two, and none of that is in the clip. The cut points are chosen on word boundaries from a transcript, not by listening for a pause.
The illustration on "less than one percent" isn't a stock overlay dropped on top — it's colour-matched to the palette of the surrounding frames, grained to the noise level measured off the source footage, and given a slow 1.5% drift so it doesn't read as a pasted-in still. The music holds level, then cuts to near-silence just before that line and comes back right after — the pause does the work, not a volume swell.
HOW IT'S BUILT
The same moment as the clip above — not a different one. Up there it's spliced from three places in the conversation; down here it's the same "humans will be 1% of all intelligence" line taken as a single unbroken take instead of a splice, so you can watch each layer get added on top of real, continuous footage and see exactly what it contributes, instead of just being told.
Same source clip in every step below — Elon Musk, on the Cheeky Pint x Dwarkesh Podcast, mid-thought ("But let me just say the total amount of intelligence... I think maybe in five or six years..."), not a warm-up line. Each step keeps everything from the one before it and adds exactly one thing — press play on any of them, or go left to right and the whole strip runs about two minutes seven seconds. The line under each clip is what it adds; the line under that is what it's actually doing for the person watching.
0
What I started from. Horizontal, no captions, no crop, nothing added.
Everything below is measured against this frame — nothing here is working for attention yet.
1
Word-by-word captions burned in, grouped by clause and punctuation. Still horizontal — nothing else changed yet.
Most people watch with sound off — the caption is often the only thing that lands.
2
Cropped to vertical (9:16). The crop point is fixed here — tracking comes next.
Craft, not the point of it — on a phone, horizontal means a rotate or black bars, an easy reason to swipe past.
3
The crop locks onto his face instead of the geometric center of frame 2.
Still craft — a fixed crop clips him when he moves, and that reads as a mistake, not a style.
4
This is the one that matters most: deciding which words actually make the cut — the sentence that carries the idea, not just whichever take came out smoothest to say.
Everything else on this page is craft on top of this decision — reframing, tracking, music and type mean nothing if the words underneath aren't the ones worth keeping.
5
Five crop-scale jumps built by hand on this one unbroken shot — medium, close, extreme close, back to medium, close again — timed to where the thought turns, not to a timer.
Craft on top of step 4's cut — a single unmoving frame reads as "phone on a stand", deliberate scale changes read as a real edit, even though the camera never cut.
6
The track isn't pulled from a library — it's generated fresh for this clip's genre and mood, then mixed with lifts at the moments that carry meaning and a hard drop to near-silence half a second before the key line. That drop measures about a 58dB fall, against roughly 3dB for an ordinary volume swell — closer to a cut than a nudge.
Craft again — silence under a talking head can feel like the clip stalled, the bed track keeps the energy up without touching a word of what's said.
7
The illustration isn't a stock graphic — it's generated specifically for what this line is saying, then colour- and grain-matched to the shot and given a slow drift so it sits in the frame instead of reading as a sticker pasted on top. The key phrase still gets large type alongside it, timed to the same beat.
The last layer of craft — type big enough to read at a glance gives the eye a beat to land on exactly where the sentence peaks.
Step 7, said plainly: this is what "effects" means in my pipeline today — the key phrase in large type, and one illustration generated for that exact line, then colour- and grain-matched into the shot. Not a stock icon library, not a transition pack — one picture, built for this sentence, repeated the same way every time.
Source clip: "Elon Musk – 'In 36 months, the cheapest place to put AI will be space'" — Cheeky Pint × Dwarkesh Podcast (youtu.be/BYXbuik3dgA). Public interview, used here to show the edit — the speaker and channel aren't affiliated with this service and don't endorse it.
WHY THIS GETS MEASURED
This isn't about being careful for its own sake — a caption that drifts or audio that clips is a clip that comes back for a redo, and a redo is days you don't get back. So the same two checks run on every edit before it ships, not eyeballed, not a one-off for this page. Ask any editor for their own numbers on these two things — most won't have an answer, because most don't measure.
style_probe.py — samples the caption's bounding box
across 40 frames of the rendered clip and logs its Y-position in
pixels.ffmpeg -af loudnorm=print_format=summary — reports
integrated loudness (LUFS) and true peak (dBTP) for the finished
mix.Competitor left un-named: the numbers are checkable against the method above, a name isn't the point.
Everything above came off the same line you'd be sending your footage to — same crop, same captions, same checks, same 48 hours. No showreel edit, no one-off polish for this page.
Send your own footage and I edit your first clip at no cost, before you pay for anything.
Send footage →If what comes back isn't right, say so — I revise it, I don't hand it back and call it done.