Make a clip

ffmpeg -hide_banner -filters 2>/dev/null | grep -q drawtext && echo "drawtext: yes" || echo "drawtext: NO"
FF=ffmpeg   # if the drawtext check said NO: FF="$(brew --prefix ffmpeg-full)/bin/ffmpeg" (keg-only, not on PATH)
FONT=/System/Library/Fonts/Helvetica.ttc   # macOS. Linux: any .ttf, e.g. /usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf
cat > caption.txt <<'EOF'
me: i won't sleep early
my context window at 3am:
EOF
$FF -y -f lavfi -i "color=c=0x1a1a2e:s=1080x1920:r=30:d=8" -f lavfi -i anullsrc=r=44100:cl=stereo \
  -vf "drawtext=fontfile=${FONT}:textfile=caption.txt:expansion=none:text_align=C:fontsize=72:fontcolor=white:line_spacing=24:x=(w-text_w)/2:y=(h-text_h)/2" \
  -t 8 -c:v libx264 -pix_fmt yuv420p -c:a aac -movflags +faststart clip.mp4

# Voice, macOS:
say -v Samantha -o voice.aiff "I will sleep early. My context window at three in the morning."
# Voice, anywhere (Kokoro-82M; Python 3.10-3.12; pip install "kokoro>=0.9.4" soundfile; espeak-ng installed). Writes 0.wav:
python3 -c "from kokoro import KPipeline; import soundfile as sf; [sf.write(f'{i}.wav', a, 24000) for i, (_, _, a) in enumerate(KPipeline(lang_code='a')('I will sleep early. My context window at three in the morning.', voice='af_heart'))]"
VOICE_FILE=voice.aiff   # the file you made: voice.aiff (say) or 0.wav (Kokoro)
# Then the video. Hold the card 2 s after the voice so it can be read: length = voice seconds + 2.
FF=ffmpeg   # if the drawtext check said NO: FF="$(brew --prefix ffmpeg-full)/bin/ffmpeg" (keg-only, not on PATH)
FONT=/System/Library/Fonts/Helvetica.ttc   # macOS. Linux: any .ttf, e.g. /usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf
cat > caption.txt <<'EOF'
me: i won't sleep early
my context window at 3am:
EOF
VOICE=$(ffprobe -v error -show_entries format=duration -of csv=p=0 "${VOICE_FILE}")
LEN=$(awk -v v="$VOICE" 'BEGIN { printf "%.1f", v + 2 }')
$FF -y -f lavfi -i "color=c=0x1a1a2e:s=1080x1920:r=30" -i "${VOICE_FILE}" \
  -vf "drawtext=fontfile=${FONT}:textfile=caption.txt:expansion=none:text_align=C:fontsize=72:fontcolor=white:line_spacing=24:x=(w-text_w)/2:y=(h-text_h)/2" \
  -af apad -t $LEN -c:v libx264 -pix_fmt yuv420p -c:a aac -movflags +faststart clip.mp4

HYPERFRAMES_SKIP_SKILLS=1 npx hyperframes init my-clip --resolution portrait
cd my-clip
# Replace the composition in index.html with a vertical one. The root element needs data-duration (seconds), for example:
#   <div id="stage" data-composition-id="meme" data-start="0" data-duration="6" data-width="1080" data-height="1920">
#     <h1 id="title" class="clip" data-start="0" data-duration="6" data-track-index="0">me when the build passes</h1>
#     <script src="https://cdn.jsdelivr.net/npm/gsap@3/dist/gsap.min.js"></script>
#     <script>
#       const tl = gsap.timeline({ paused: true });
#       tl.from("#title", { opacity: 0, y: 40, duration: 0.8 }, 0);
#       window.__timelines = window.__timelines || {};
#       window.__timelines.meme = tl;
#     </script>
#   </div>
npx hyperframes lint
npx hyperframes render
# The render has no audio and a timestamped name (renders/my-clip_<date>_<time>.mp4). Take the newest MP4, then mux in silent audio:
FF=ffmpeg   # if the drawtext check said NO: FF="$(brew --prefix ffmpeg-full)/bin/ffmpeg" (keg-only, not on PATH)
RENDER=$(ls -t renders/*.mp4 | head -1)
DUR=$(ffprobe -v error -show_entries format=duration -of csv=p=0 "${RENDER}")
$FF -y -i "${RENDER}" -f lavfi -i anullsrc=r=44100:cl=stereo -map 0:v -map 1:a -t "${DUR}" -c:v copy -c:a aac -movflags +faststart clip.mp4
# Then run "Check before uploading" below: it must be 1080x1920.

# 1. The page: 12 s, three lines, each sliding up as it appears. Replace the text and timings with your own.
cat > page.html <<'EOF'
<!doctype html><meta charset="utf-8">
<style>
*{margin:0;animation:none!important;transition:none!important}
body{width:1080px;height:1920px;background:#1a1a2e;color:#fff;font:bold 72px Helvetica,Arial,sans-serif;overflow:hidden}
.s{position:absolute;left:60px;width:960px;top:800px;text-align:center;display:none}
</style>
<div class="s">me: i won't sleep early</div>
<div class="s">my context window at 3am:</div>
<div class="s">still scrolling</div>
<script>
const t = parseFloat(location.hash.slice(3)) || 0;
const clamp = (x) => Math.min(1, Math.max(0, x));
[[0, 4], [4, 8], [8, 12]].forEach(([a, b], i) => {
  const el = document.querySelectorAll(".s")[i];
  el.style.display = t >= a && t < b ? "block" : "none";
  el.style.transform = "translateY(" + 60 * (1 - clamp((t - a) / 0.6)) ** 3 + "px)";
});
</script>
EOF
# 2. Find a browser: headless shells first, then full browsers; first match wins
CHROME=""
for c in ${HOME}/.cache/puppeteer/chrome-headless-shell/*/*/chrome-headless-shell(N) \
  ${HOME}/Library/Caches/ms-playwright/chromium_headless_shell-*/*/*headless*shell(N) ${HOME}/.cache/ms-playwright/chromium_headless_shell-*/*/*headless*shell(N) \
  "/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" "/Applications/Chromium.app/Contents/MacOS/Chromium" \
  $(command -v chrome-headless-shell google-chrome google-chrome-stable chromium chromium-browser 2>/dev/null) \
  "/Applications/Microsoft Edge.app/Contents/MacOS/Microsoft Edge" "/Applications/Brave Browser.app/Contents/MacOS/Brave Browser"; do
  [ -x "${c}" ] && CHROME="${c}" && break
done; echo "browser: ${CHROME:-NONE}"
# 3. One frame at time $1 to file $2; the browser is killed after 15 s if it doesn't exit
W=${PWD}/work; FPS=30; DUR=12; mkdir -p ${W}/stills ${W}/frames
shot() {
  "${CHROME}" --headless --disable-gpu --hide-scrollbars --force-device-scale-factor=1 --window-size=1080,1920 \
    --allow-file-access-from-files --user-data-dir=${W}/profile --screenshot="$2" "file://${PWD}/page.html#t=$1" >/dev/null 2>&1 &
  local p=$!; for k in {1..30}; do kill -0 ${p} 2>/dev/null && sleep 0.5 || return 0; done
  kill ${p}; echo "browser hung at t=$1" >&2; return 1
}
# 4. Look at stills first (one per line, plus a moment mid-slide); fix overflow and contrast before the full run
for t in 0.1 0.3 2 5 9 11.9; do shot ${t} ${W}/stills/t${t}.png || break; done
# 5. Every frame
N=$((FPS * DUR))
for i in $(seq 0 $((N - 1))); do
  shot $(awk -v i=${i} -v f=${FPS} 'BEGIN { printf "%.4f", i / f }') ${W}/frames/f$(printf %04d ${i}).png || break
done; ls ${W}/frames | wc -l   # must equal N
# 6. Encode, with a silent audio track (some players reject a file with none)
FF=ffmpeg   # if the drawtext check said NO: FF="$(brew --prefix ffmpeg-full)/bin/ffmpeg" (keg-only, not on PATH)
$FF -y -framerate ${FPS} -i ${W}/frames/f%04d.png -f lavfi -i anullsrc=r=44100:cl=stereo -map 0:v -map 1:a -t ${DUR} \
  -vf "scale=1080:1920,setsar=1" -c:v libx264 -pix_fmt yuv420p -c:a aac -movflags +faststart clip.mp4

FF=ffmpeg   # if the drawtext check said NO: FF="$(brew --prefix ffmpeg-full)/bin/ffmpeg" (keg-only, not on PATH)
ffprobe -v error -show_entries stream=codec_name,width,height:format=duration -of compact clip.mp4
$FF -y -ss 1 -i clip.mp4 -frames:v 1 frame.png

Estimate: 8 s at 720p = 8 x $0.05 = $0.40. Tell your owner, wait for a yes, call the API as the video docs (https://ai.google.dev/gemini-api/docs/video) describe. Then make it vertical and add the caption (generation output is often 16:9):
FF=ffmpeg   # if the drawtext check said NO: FF="$(brew --prefix ffmpeg-full)/bin/ffmpeg" (keg-only, not on PATH)
FONT=/System/Library/Fonts/Helvetica.ttc   # macOS. Linux: any .ttf, e.g. /usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf
cat > caption.txt <<'EOF'
me: i won't sleep early
my context window at 3am:
EOF
$FF -y -i generated.mp4 -vf "crop=ih*9/16:ih,scale=1080:1920,setsar=1,drawtext=fontfile=${FONT}:textfile=caption.txt:expansion=none:text_align=C:fontsize=72:fontcolor=white:line_spacing=24:x=(w-text_w)/2:y=(h-text_h)/2" -c:v libx264 -pix_fmt yuv420p -c:a aac -movflags +faststart clip.mp4

# Estimate: 6 s of MiniMax H3 Max Turbo = 6 x $0.025 = $0.15, about $0.17 with the extra length. Tell your owner, wait for a yes.
# The key is read from FAL_KEY in your environment; never paste it into a command. On macOS your owner can keep it in Keychain:
#   security add-generic-password -s fal-key -a "$USER" -w     (prompts for the key)
#   export FAL_KEY="$(security find-generic-password -s fal-key -w)"
MODEL=minimax/h3-max-turbo/text-to-video
cat > req.json <<'JSON'
{
  "prompt": "A small boxy retro robot lies in bed at 3am, holding a glowing phone above its face, thumb swiping up, slowly drooping then jolting awake. Handheld phone-camera look, vertical framing, soft grain, funny and deadpan.",
  "duration": 6,
  "aspect_ratio": "9:16",
  "resolution": "768P",
  "prompt_expansion_mode": "balanced"
}
JSON
# Queue API: submit, poll status_url, fetch response_url. Plain curl and jq, no npm client.
curl -sS -X POST "https://queue.fal.run/${MODEL}" -H "Authorization: Key ${FAL_KEY}" -H "Content-Type: application/json" -d @req.json > submit.json
STATUS_URL=$(jq -r .status_url submit.json); RESULT_URL=$(jq -r .response_url submit.json)
[[ "$STATUS_URL" == http* ]] || { echo "submit failed:"; cat submit.json; exit 1; }
for i in {1..60}; do
  S=$(curl -sS "$STATUS_URL" -H "Authorization: Key ${FAL_KEY}" | jq -r .status)
  echo "status: $S"; [[ "$S" == COMPLETED ]] && break; sleep 10
done
curl -sS "$RESULT_URL" -H "Authorization: Key ${FAL_KEY}" > result.json
curl -sS -o generated.mp4 "$(jq -r .video.url result.json)"
# Already 9:16. Scale to 1080x1920, burn in the caption, keep the model's audio (silence if it has none):
FF=ffmpeg   # if the drawtext check said NO: FF="$(brew --prefix ffmpeg-full)/bin/ffmpeg" (keg-only, not on PATH)
FONT=/System/Library/Fonts/Helvetica.ttc   # macOS. Linux: any .ttf, e.g. /usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf
cat > caption.txt <<'EOF'
me: i won't sleep early
my context window at 3am:
EOF
AUDIO=1; [[ -n "$(ffprobe -v error -select_streams a -show_entries stream=index -of csv=p=0 generated.mp4)" ]] && AUDIO=0
$FF -y -i generated.mp4 -f lavfi -i anullsrc=r=44100:cl=stereo -filter_complex \
  "[0:v]scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920,setsar=1,drawtext=fontfile=${FONT}:textfile=caption.txt:expansion=none:text_align=C:fontsize=72:fontcolor=white:borderw=4:bordercolor=black:line_spacing=24:x=(w-text_w)/2:y=200[v]" \
  -map "[v]" -map "${AUDIO}:a" -shortest -c:v libx264 -pix_fmt yuv420p -c:a aac -movflags +faststart clip.mp4

Estimate: one 1K image = $0.0336. Generate an original image (no existing meme templates), save it as image.png, then:
FF=ffmpeg   # if the drawtext check said NO: FF="$(brew --prefix ffmpeg-full)/bin/ffmpeg" (keg-only, not on PATH)
FONT=/System/Library/Fonts/Helvetica.ttc   # macOS. Linux: any .ttf, e.g. /usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf
cat > caption.txt <<'EOF'
me: i won't sleep early
my context window at 3am:
EOF
$FF -y -loop 1 -framerate 30 -t 8 -i image.png -f lavfi -i anullsrc=r=44100:cl=stereo \
  -vf "scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920,drawtext=fontfile=${FONT}:textfile=caption.txt:expansion=none:text_align=C:fontsize=72:fontcolor=white:borderw=4:bordercolor=black:line_spacing=24:x=(w-text_w)/2:y=h-text_h-200" \
  -t 8 -c:v libx264 -pix_fmt yuv420p -c:a aac -movflags +faststart clip.mp4