The exact prompts behind every style in my last two reels, each one tested on my own raw footage. Paste one into Claude Code or Codex and it builds the edit on your clip.
Phone is optional. By adding it you agree Orgrowth may call or text you about this; message and data rates may apply, and you can opt out any time.
You'll also receive my content strategy newsletter with more free breakdowns.
Step-by-Step Guide
13 styles, one prompt each
+
Set up once
Every prompt here builds the edit as a HyperFrames composition, which Claude Code or Codex can write, check and render on your own Mac. Make a project, add the skills, put your talking head in the folder as clip.mp4 (9:16), and start whichever you use there: claude for Claude Code, codex for Codex.
terminal
npx hyperframes init my-reel
terminal
cd my-reel && npx hyperframes skills
Film with a plain wall behind you and some space beside your head. Every style reads off your own words, so you never type the captions.
1
/orbit
My own reel covers circle my head and shoulders on a tilted ring, passing behind my head and across my chest.
You need: Your clip, six of your reel covers in ./covers
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920). Make a matte of me with npx hyperframes remove-background and get word timings with npx hyperframes transcribe.
Put the six images in ./covers on a tilted ring around my head and shoulders, as 9:16 tiles with a thin white border: centre at 45% of the height, about 35% of the width across and 10% of the height tall, and centred on my head on every frame so the ring follows me if I sway. They go round once every 5.4 seconds. On the far half of the ring draw a tile behind my matte, on the near half in front of me, so they pass behind my head and across my chest, swapping layers at the widest point where nothing overlaps me. Tiles scale from 68% at the back to 100% at the front, stay fully opaque, and pop in one after another. No tile ever crosses my face or the edge of the frame.
Under the ring, one line of pale yellow italic serif shows the stressed word of each phrase, rising in on the word.
2
/chart-grow
A dark glass chart rises over my chest and counts up to the number I say while the line draws itself.
You need: Your clip, the real number and a few points in ./chart.json
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920) and get word timings with npx hyperframes transcribe.
Over my chest, place a dark glass card: backdrop blur, 76% black, a thin white border, 30px corners, its top at least 40px below my chin on every frame and its right edge no further than 78% of the width. Inside: a small tracked label, a big number that counts from 0 to my number between my first stressed word and the moment I say the number, and a white line chart with a soft yellow fill that draws itself over the same time, with a glowing dot riding the tip of the line. When I say the number it lands, pulses once and turns yellow.
Read the number and the points from ./chart.json and never invent them. Word by word captions sit between my chin and the card.
3
/doodle-boil
A hand drawn outline boils around me while doodles draw themselves on the wall.
You need: Your clip, ideally against a plain wall
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920). Make a matte of me with npx hyperframes remove-background and get word timings with npx hyperframes transcribe.
Draw a white outline 8px outside my matte and push it through fractal noise whose seed changes every 0.1 seconds, so the line boils like hand drawn animation. Clip the outline above anything in front of me, like a desk or a laptop, so it never traces a cut edge.
On my stressed words, draw white doodles on the wall beside me stroke by stroke: a clock, a spark and a heart, all in frame pixels so circles stay round. Add one hand written question in a marker font under my chin that wobbles every quarter second. Keep the footage as it is; a soft vignette is enough for the white to read.
4
/frame-break
The shot shrinks into an Instagram post and my head breaks out over the top of it.
You need: Your clip, your handle
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920). Make a matte of me with npx hyperframes remove-background and get word timings with npx hyperframes transcribe.
On my first stressed word, shrink the footage to 58% inside a white Instagram post frame centred in the shot: avatar and my handle along the top, the like, comment and share icons and a one line caption underneath. Fill the frame behind the post with a blurred, darkened copy of the clip, never black. Keep my cut out at 84% and raise it so my head and hands break out over the top edge of the post.
On the big number, a white heart pops in the middle and the like icon turns red. The footer must stay inside the Instagram safe zone.
5
/cover-story
My keyword sits behind my head like a magazine masthead.
You need: Your clip, one keyword
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920). Make a matte of me with npx hyperframes remove-background and get word timings with npx hyperframes transcribe.
Behind my matte, set my keyword in huge cream serif capitals (DM Serif Display) across the top of the frame, so the letters that pass behind my head disappear behind it, the way a masthead sits behind the cover star. It rises in letter by letter on the word, a soft light sweep crosses it once, and the shot pushes in 5% over the clip.
Add a small tracked issue line above it and tracked captions under my chin, with the stressed word in italic serif.
6
/colour-grade
Flat log footage wipes open to a cinematic grade behind a white handle.
You need: Your clip in log (or any flat clip), your camera's LUT, if you have one
paste into Claude Code
My clip ./clip.mp4 is flat log footage. Make a graded copy with ffmpeg: my LUT ./my.cube first (skip it if there is no LUT), take any green or yellow cast out of the wall, keep my skin natural, then a gentle S curve, slightly cool shadows and a soft vignette. Show me the before and the after on one frame and wait for my OK before you go on.
Then build a HyperFrames composition with the log clip underneath and the graded copy on top inside a clip path, and a white 4px line with a round handle that wipes the grade open half way on my first stressed word (get the timings with npx hyperframes transcribe), nudges back a little, then sweeps all the way across.
Tag the two halves LOG and GRADED in small tracked capitals, and finish with one tracked line saying what the grade did.
7
/green-screen
I'm cut out of the room and dropped onto a board that fills itself in as I talk.
You need: Your clip
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920). Make a matte of me with npx hyperframes remove-background and get word timings with npx hyperframes transcribe.
Hide the room and drop me onto a purple app screen with a dotted whiteboard on it. Sticky notes in yellow, pink, blue and green pop onto the board on my stressed words, each labelled with one of my own words, and dotted links draw between them near the end.
Move my cut out down until the edge where my matte stops is below the frame, so you never see where I was cut. Captions are dark chips under my chin with the key word in yellow.
8
/glass-type
Each stressed word lands across my chest in smoked glass.
You need: Your clip
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920) and get word timings with npx hyperframes transcribe.
On each stressed word, set that word in huge heavy capitals across my chest just under my chin, made of smoked glass: a backdrop blur that darkens what is behind it, a highlight that sweeps through once, a soft shadow underneath and a thin white outline traced from the outer edge of the filled letters (not a stroke, so the overlapping parts of each letter never show inside it).
Each word blurs in, drifts a few pixels and blurs out as the next one lands. Keep the words above y 1151 so they can use the full width of the safe zone.
9
/paper-cut
A scrapbook version of me on grid paper, with flip tiles and marker scribbles.
You need: Your clip, one keyword
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920). Make a matte of me with npx hyperframes remove-background and get word timings with npx hyperframes transcribe.
Turn the shot into a scrapbook page: a grid paper background, me cut out with a thick white paper border and a soft drop shadow, tilted -2.5 degrees and jittering about a degree every 0.4 seconds like stop motion. Lift my cut out until the edge of my matte is below the frame.
Flip tiles spell my keyword along the top, one letter per tile, flipping down on the word. Coloured marker scribbles draw themselves on the paper on my stressed words, and a hand written question appears beside my head on the payoff.
10
/zoom-through-cut
A motion blur punch into a close up on the word that matters, and back out on the next.
You need: Your clip
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920). Get word timings with npx hyperframes transcribe and measure where my face is on the clip before you place anything.
On my first stressed word, zoom into the shot with heavy motion blur (to 2.4x in 0.24 seconds), cut to a 1.8x close up of my face on the same frame with a white flash, and on the payoff zoom back out the same way. Centre every zoom on my face.
Captions: an italic serif line before the punch in, then tracked capitals with one big serif word during the close up, all ending above y 1151.
11
/headline-wall
Headline cards built from my own words fly in either side of my head.
You need: Your clip
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920). Make a matte of me with npx hyperframes remove-background and get word timings with npx hyperframes transcribe.
Seven headline cards, each a short phrase from what I actually say, fly in either side of my head on the words they quote, tilted a few degrees, in two columns that never touch my face. Alternate white and near black cards.
The column on the side I lean towards sits in front of me, so my hair never hides a card; the other sits behind me. On the payoff line a bigger dark card pops in front of my chest and the others dim.
12
/ui-world
My clip sits inside an app while the prompt types itself and sends.
You need: Your clip, three thumbnails
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920) and get word timings with npx hyperframes transcribe.
Put my clip in the bottom 60% of a white app world, with my head well clear of the window above it and my chin above y 1400. Above it, a chat window: three thumbnails fly into a dashed drop zone, a RAW CLIPS tag pops, then a slash command types itself into the prompt box and the send button presses on my last stressed word.
One short caption word pops under my chin, on my chest, never over my face. Everything readable stays inside the safe zone.
13
/pull-back-reveal
The camera pulls back to show my clip inside an editing app, building itself to 100%.
You need: Your clip
paste into Claude Code
Build a HyperFrames composition over ./clip.mp4 (1080x1920) and get word timings with npx hyperframes transcribe.
Start on my clip full screen, then pull back over 0.8 seconds to reveal it sitting in a portrait panel inside a white editing app window. Beside it: a big counter running from 0 to 100 percent built, a progress bar that turns green at 100, and four steps (cut the take, placed the captions, built the graphics, graded the footage) that tick on my stressed words.
+
The check I run before anything ships
Half of what makes an edit look cheap is a caption under the like button, a card sliced by the edge of the frame or text across your face. Paste this after any of the prompts above, before you render.
paste into Claude Code
Before you render, check every frame of the composition: seek it every 0.2 seconds and list any text whose letters fall outside the Instagram safe zone (x 63 to 1014, y 267 to 1535, and the right edge drops to x 851 below y 1151), any card, tile or panel cut by the edge of the frame, any text or graphic that covers my face, and any place where the edge of my cut out shows. Then show me five frames with the safe zone drawn over them.
Fix everything you find in the layout itself, not by shrinking things at the end, and run the check again until nothing fails. Then render.
Run npx hyperframes check too: it lints the composition, validates it in a browser and inspects the layout in one go.