Our video-overlay/captions path renders static PNGs composited by ffmpeg. When a job
needs real motion design — eased entrances, staggered reveals, word-synced kinetic
captions, animated lower-thirds, branded title cards, count-ups — author the graphic as a
Remotion React component (every word/color/beat is a prop, animated over frame), render
it to a transparent layer, then composite with our existing ffmpeg overlay. ffmpeg still
owns the cut, audio, grade, and final encode; Remotion only authors the moving graphic.
⚠ Before using — it's optional and heavy
- Needs Node + npm and a Remotion project (the
remotion/scaffold in this repo). First render downloads a headless Chromium (~150–300 MB). This is a bigger dependency than the whole Python core — like thegrok-video-editvfx extra, it's off by default. - Licensing: Remotion is free for individuals/small teams but companies above a small headcount need a paid seat license — verify before using under a company. (No API key, but a real license gate.)
- Render cost: frame-by-frame Chrome rendering; render the overlay at overlay size/duration only (seconds of a lower-third), never the whole video, to keep it fast.
- If the graphic is static or a simple fade → use
video-overlay, not this.
Default pattern: transparent layer → ffmpeg composite
- Author/parameterize the component (see scaffold in
remotion/src/). Pass content + timing asinputPropsJSON — no hard-coded text. - Render only the graphic on alpha:
cd remotion && npm install # first time only (pulls Remotion + Chromium) npx remotion render LowerThird out/lower.mov \ --codec=prores --prores-profile=4444 --pixel-format=yuva444p10le \ --props='{"name":"Lior Kolnik","title":"Product, Eve Security","inSec":0.3,"outSec":4.0}' # (or a transparent WebM: --codec=vp8 --pixel-format=yuva420p) - Composite over the footage with our normal overlay (timed window), keeping audio:
This is exactly the RGBA-sequence compositing inffmpeg -y -i footage.mp4 -i remotion/out/lower.mov -filter_complex \ "[0:v][1:v]overlay=40:H-h-60:enable='between(t,12,16)'" \ -c:v h264_videotoolbox -b:v 9M -c:a copy out.mp4video-overlay— Remotion just produced a richer, animated layer. For a full-frame animated title/outro card (no transparency), render opaque andconcatit between clips (or drop it into anedl-editEDL as a clip).
Word-synced timing comes from the transcript (cue sheet)
Don't eyeball when a word/overlay lands — grep the --words transcript for the phrase,
read its timestamp, convert to a frame (frame = round(t * fps)), and put it in the cue sheet.
The scaffold's FinalEdit.tsx holds CUES = [{id, at, dur}]; e.g. an emphasis lands on the
word "right" at 4.92s → at: Math.round(4.92*fps). Kinetic captions read the same word
timestamps to pop one word per beat. (See editor's "the edit is text" convention.)
Tuning knobs + reviewing your own work
- Global feel lives in
remotion/src/anim.ts— a fewTIMINGconstants (reveal, stagger, overlayIn/Out) + a shared easing. "Make it snappier" = change one number, re-render. - Review before a full pass:
npx remotion still <Comp> still.png --frame=N --props=...renders one frame — read it (like our frame-extraction checks), adjust, then do the fullrender. Iterate stills, not full renders.
When to reach for it (vs not)
- Use Remotion: kinetic/karaoke captions, animated lower-thirds, branded animated intro/ outro cards, count-ups/charts, anything with eased motion.
- Use
video-overlay/captionsinstead: a static logo/label/watermark, burned-in plain captions, a simple fade-in PNG — cheaper, no Node/Chromium, no license question.
See remotion/README.md for the scaffold layout and adding a new component.