Zoom punch-ins, cards behind your shoulders, a glass graph, text behind your head and a tier list. Claude builds each one on your clip and renders an MP4. Then you turn them into one-word commands you can run from your phone.
You don't write any code. Claude writes it with Remotion, a free tool that turns code into video. You paste a prompt, look at the stills it shows you, and say what to change.
Run these 2 prompts one time. The first makes the project. The second cuts you out of the background, which 2 of the effects need.
Set up a Remotion video project in a folder called claude-edits on my Desktop. Install everything it needs.
Rules for every video in this project:
- 1080 wide, 1920 tall, 30 frames per second.
- Keep all text and graphics between x=120 and x=960, so Instagram's buttons and caption never cover them.
Copy my clip from [PATH TO YOUR CLIP] into the project's public folder as talking.mp4. If it's iPhone HDR footage, convert it to normal colour first so it doesn't look grey. Never change my original file.
Every time I ask for an effect:
1. Build it as its own composition, named after the effect.
2. Open Remotion Studio so I can preview it.
3. Render it to an MP4 in a folder called out and tell me the exact path.
4. Show me 3 or 4 still frames from it.
Save these rules in the project's CLAUDE.md so you follow them in every chat.To get the path of your clip on a Mac, drag the file into the Claude message box.
Make a person mask for public/talking.mp4: white where I am, black everywhere else, one image per frame, same size and frame rate as the clip. Save it in public/mask.
On a Mac, use Apple's Vision person segmentation. On Windows, use rembg.
Smooth the edges a little and stop it flickering between frames. If it picks up a lamp, a light or a mic next to me, clean that out.
Show me 5 frames of me cut out over a green background so I can check it.
Then make a reusable component called BehindMe that plays my clip with any graphic drawn behind me and in front of the background.This takes a few minutes. About 7 for a 4-second clip on my Mac. Look at the green frames. Lights and mics next to you are the usual mistakes. If you see one, tell Claude what it is and where.
The camera snaps in on your face on your key words. Each punch is tighter than the last.
Build a zoom punch-in effect on my clip.
Transcribe the clip with word timings and snap the camera in on my face on [2 OR 3 WORDS, for example "money", "free", "now"].
Each punch is tighter than the last: 1.15x, then 1.3x, then 1.45x. Fast in with a small bounce, with a little motion blur on the snap.
Keep my face centred and never show black edges.Two dark glass cards sit behind your shoulders. The camera pushes over one shoulder for BEFORE, swings across for AFTER, then settles back.
Build an over-the-shoulder effect, 84 frames.
Put two dark glass cards behind me, just above my shoulders, one on each side. Use BehindMe so my head and shoulders stay in front of them. Measure my shoulders from the mask to place them.
Left card: BEFORE, [YOUR NUMBER, for example 1,204 views].
Right card: AFTER, [YOUR NUMBER, for example 412K views].
Camera: start wide, push in 1.3x over my left shoulder as the BEFORE card pops in, swing across and push in over my right shoulder as AFTER pops in, then settle back to the full shot.
Ease every move. Add motion blur that grows with camera speed. Never show black edges. Keep the card in focus inside x=120 to 960.
Show me stills at the start, the left push, the right push and the end.A frosted glass card under your face. A big number counts up while an orange line graph draws itself, and the dot on the tip pulses when it lands.
Build a glass graph effect, 90 frames.
A frosted glass card at chest height, under my face. Measure where my chin goes over the clip so the card never covers my face.
Inside the card: a small grey title, [YOUR TITLE, for example Views this month], then a big white number counting from [START, for example 1,204] to [END, for example 412K], then an orange line graph that draws itself left to right with a soft glow, a fill fading down, and a glowing dot on the tip that pulses once at the end.
Count through the orders of magnitude so the low numbers show, not a jump straight to the end.
The card slides up and fades in, holds, then fades out.
Show me stills at frames 10, 40 and 70.A huge word rises up behind you. Your hair covers part of it, so it reads as sitting behind your head.
Build a text-behind-my-head effect, 90 frames.
A huge bold white word, [YOUR WORD, for example VIRAL], behind my head and in front of the background. Use BehindMe.
Use a heavy condensed font already on my computer. Fit the whole word between x=120 and x=960, and shrink it if it's too wide.
Place it at the top of my head, so my hair covers the bottom middle of the letters but the word still reads. Measure my head from the mask.
It rises in from blurry to sharp, drifts up slowly, then drops and fades out.
Show me stills at frames 5, 40 and 85.A dark S to D tier list under your face. Items drop into their rows one by one, and the status pill flips from Ranking to Ranked.
Build a tier list effect, 90 frames.
A dark panel under my face with 5 rows, S, A, B, C and D, each with its own colour. Measure my chin so the panel never covers my face.
Topic: [YOUR TOPIC, for example AI tools].
S: [ITEMS]. A: [ITEMS]. B: [ITEMS]. C: [ITEMS]. D: [ITEMS].
Items drop into their rows one at a time, every 4 frames, with a small bounce.
If there's a logo in public/logos named after an item, use it. Otherwise show the name as a text chip.
A pill at the top says "Ranking..." with a spinner, then changes to "Ranked" with a tick when the last item lands.
Show me stills at frames 15, 45 and 75.For logos, save each one as a PNG named after the item, for example public/logos/claude.png.
Once you like an effect, make it a one-word command. Then you can type /glass-graph 900 → 2.1M and get the video back, with no long prompt.
Turn each effect we built into its own personal skill, so it works as a slash command in any chat: /zoom-punch-in, /over-the-shoulder, /glass-graph, /text-behind and /tier-list.
Each one reads what I type after it (a word, two numbers, a topic and items), renders that effect on my clip without sound, and sends me the MP4 in the chat with a one-line reply. No plan, no long explanation.
If I attach a new clip, set it up first: convert a copy into the project, make its mask, and measure my face and shoulders. Then render.| Type | You get |
|---|---|
| /zoom-punch-in | 3 punches on your face |
| /over-the-shoulder 1,204 → 412K | BEFORE and AFTER cards behind you |
| /glass-graph 1,204 → 412K | The glass card counting up |
| /text-behind RICH | RICH behind your head |
| /tier-list AI tools | The tier list, filled in |
Each one comes back in about 20 to 40 seconds.
This is the part people see in the video: you type a command on your phone and the edited clip comes back. Your Mac does the work. Your phone is the remote.
Keep the Mac awake with the Claude app open the whole time you film. If it sleeps, the command never runs.
In your editor, cut out the waiting, put the result right after the command, and add music.
Remotion is free for individuals and very small teams. Bigger companies need a licence. Check remotion.dev/license before you use it for client work.