Free Video Effect
Digital Zoom Effect: Punch-Ins on the Words That Matter
The shot punches in on your face on key words, then eases back, with bold two-colour captions.
Claude Code · Opus 5.59:16Easy
- Run the setup once
- Paste the prompt, fill in the [brackets]
- Approve the preview, get your MP4
New to Claude Code? Start with the beginner guide →
Want all 17 at once? Get the full pack →
The prompt
effects/digital_zoom.py: add a "Digital Zoom" effect to the whole video.
- Read words.json and pick the [6-8] most important words for the meaning of the video yourself
(or use these words: [WORD 1, WORD 2, ...]).
- On every accent word, zoom in toward my face (find the face center with MediaPipe Face
Detection):
- scale 1.00 → 1.18 in 120 ms, easeOutCubic (punch-in)
- when the word ends, ease back to 1.06 within 400 ms
- alternate a close (1.18) and a medium (1.10) zoom so it doesn't repeat
- A light 2-frame motion blur during the zoom.
- Captions: in the lower third, in word groups (2-3 words). The accent word is split
into two lines: top line white, bottom line [ACCENT COLOR], very bold,
with a light outer shadow. Words pop in with scale 0.85→1.04→1.
- The image must not get blurry while zooming: scale the source frame with high quality
(Lanczos).Fill in the brackets
| Placeholder | What to put there | Example |
|---|---|---|
[6-8] | How many accent words Claude should pick if you let it choose. | 6-8 |
[WORD 1, WORD 2, ...] | Optional: your own accent words, spelled as they appear in words.json. Use either this list or Claude's choice, not both. | free, prompt, setup |
[ACCENT COLOR] | Colour of the bottom caption line on accent words, as a hex code. | #FFE14D |
About Digital Zoom
Digital Zoom is a talking-head video effect that punches in towards your face on the words that matter, then eases back, with bold two-colour captions on those words. After the one-time setup, paste the prompt into Claude Code with Opus 5.5; Claude picks six to eight key words from your transcript, or uses yours, and renders the zooms from your footage.
Use it when: In every video; it is the simplest effect in the set and works as a base layer under the others.
What the Digital Zoom effect does
Digital Zoom adds punch-ins on the words that carry your point. When an accent word starts, the frame jumps from 1.00 to 1.18 scale in 120 ms, centred on your face, then settles back to 1.06 over 400 ms once the word ends. The next accent word gets a medium push to 1.10 instead, then a close one again, so the rhythm alternates rather than repeats.
Captions sit in the lower third in groups of two or three words. On an accent word the caption splits over two lines, the top one white and the bottom one in your accent colour, heavy and lightly shadowed, and each word pops in as it arrives. Two frames of motion blur on every zoom make the jump feel like a camera move rather than a resize.
When it works and when it hurts
This is the base layer. It fits almost any talking-head video because it uses only your own footage and transcript, with no extra images to prepare, which also makes it the natural first effect to try after the setup.
It hurts when it fires too often. Each punch-in tells the viewer that this word matters, and a dozen of those in thirty seconds tell them nothing. It also exposes your footage: at 1.18 the frame shows about 85% of the original width and height, stretched back to full size, so soft focus, noise or compression in the source become easier to see. And if Claude picks the words, it reads them from the transcript, not your delivery. A word can look important on paper and sound flat on camera.
How Claude builds it
Claude starts with words.json. Either it reads the transcript and chooses the words that carry the meaning, or it looks up the words you listed. Each one has a start and end time, which become frame numbers.
For the zoom centre, Claude runs MediaPipe Face Detection on the frames and takes the middle of the detected face. For every frame, the script works out the scale from the easing curves, crops a window around your face and resizes it back to 1080×1920 with Lanczos resampling, which keeps edges crisper than simpler methods. On zoom frames it blends sub-frame samples for the motion blur. Captions are drawn last, and ffmpeg writes the MP4 with your original audio.
Make it yours
The numbers are taste. A 1.12 close shot is gentler, a 1.25 one more aggressive, and a slower 200 ms push-in feels less like a cut. Choose the words yourself if you care exactly which moments land. You can also drop the captions from this effect and let another caption style run on top, or aim the zoom a little above your face so the captions never reach your chin.
Questions
Why is Digital Zoom rated Easy?
It needs no extra images, no person mask and no HTML design. It works from your transcript and the position of your face, both of which Claude gets automatically, and the motion is a single change of scale. That makes it a sensible first effect for learning how previews, notes and re-renders work.
Will zooming make my video blurry?
A little, because the close shot enlarges part of the frame. Lanczos resampling keeps the loss small, and a well-lit, in-focus source helps more than any setting. If close shots still look soft, lower the maximum scale or export at a lower CRF.
Can I pick the words myself?
Yes. Replace [WORD 1, WORD 2, ...] with your list and remove the part that asks Claude to choose. Spell each word the way it appears in words.json, and if a word comes up more than once, tell Claude which occurrence you mean, for example by its time.
Can I combine it with other effects?
Yes. It works well as the default between other effects. Run it across the whole video and ask Claude to skip the ranges where another effect is active, or use the combining prompt and list Digital Zoom for specific ranges of the timeline.
Step by step
Run the setup once
Open the folder with input.mp4 in Claude Code, choose Opus 5.5 and paste the setup prompt. Digital Zoom needs the words.json transcript it creates.
Decide who picks the words
Either leave the [6-8] count and let Claude choose, or fill in your own list and delete the part of the line that asks Claude to choose.
Paste and fill in the prompt
Copy the Digital Zoom prompt, fill in the count or your words and your accent colour, and send it.
Review the word list, then the render
Ask Claude to list the words it picked with their times before rendering. Swap any that feel wrong, then watch the result and give notes such as 'softer close shots'.
You need: Setup prompt (run once) · Word timestamps · Face detection · Sharp, well-lit footage
Tips
- The source suggests using it on every video. It needs nothing beyond the setup, so it is a good first effect for learning the preview-and-notes workflow.
- Ask Claude to list the words it picked, with their times, before it renders. Swapping a word takes one sentence; a re-render takes much longer.
- Six to eight accent words is a good starting range for a short video. If the zooms stop feeling special, reduce the number rather than the size.
- Sharp focus and good light on your face matter more here than in most effects, because the punch-in magnifies whatever is in the source.
- If you fill in your own words, delete the 'choose them yourself' part of the line so Claude does not mix both approaches.
If it goes wrong
The zoom slides sideways or misses your face when you turn your head.
Face detection lost you for a few frames. Ask Claude to smooth the face centre over time and hold the last good position whenever detection fails.
The close shots look soft.
Check that Lanczos is used for every resize and lower the close shot to 1.12. Better light, sharper focus or a lower CRF on export also help.
The zooms feel repetitive or tiring.
Ask for fewer accent words, or make the medium shot gentler (1.06–1.08) so the close shots stand out.
Captions sit on your chin during close shots.
Ask Claude to put the zoom centre slightly above the face centre, or move the caption block lower. Change nothing else.
What it does not do
- It is a crop, not a camera move. At 1.18 the picture is enlarged from about 85% of the frame's width and height, so focus, noise and compression problems in your footage become more visible.
- Face detection needs a visible, roughly front-facing face. Turning away, covering your face with a hand or leaving the frame makes the zoom centre unreliable.
- When Claude chooses the accent words, it works from the transcript. It cannot hear your emphasis, so its picks are a draft to review, not a final edit.
Sources
- Face detection guide — Google AI Edge — MediaPipe. Accessed October 21, 2026.
- ffmpeg filters documentation — FFmpeg. Accessed October 21, 2026.
- Easing functions cheat sheet — easings.net. Accessed October 21, 2026.
- faster-whisper — GitHub — SYSTRAN. Accessed October 21, 2026.
Want the whole library?
Lifetime opens every AI Video Effect, AI Motion Graphic and AI Creator Concept, inside Claude via MCP, for one payment of $79.