Instagram Reel Remix

If any install fails, stop and surface the error.

Step 1 — Download Source Video

WORK_DIR="$(pwd)/reel-remix-work"
mkdir -p "$WORK_DIR"
yt-dlp --no-playlist -o "$WORK_DIR/original.mp4" "<INSTAGRAM_URL>"

Verify it downloaded. If yt-dlp fails (geo-block, login-required, private), surface the error — do not continue.

Step 2 — Inspect the Video

ffprobe -v quiet \
  -show_entries stream=codec_name,width,height,duration \
  -show_entries format=duration \
  -of default "$WORK_DIR/original.mp4"

Extract and store:

• VIDEO_W — width (typically 720)
• VIDEO_H — height (typically 1280)
• AUDIO_DUR — format duration (the true total length; this is the master clock)

Extract 3 probe frames and auto-detect where the video content starts (= height of the text zone):

# Extract frames at positions 0, 15, 30
ffmpeg -i "$WORK_DIR/original.mp4" \
  -vf "select=eq(n\,0)+eq(n\,15)+eq(n\,30)" \
  -vsync 0 "$WORK_DIR/probe_%d.png" -y 2>/dev/null

# For each probe frame, find the y-offset where non-background content begins
# Take the minimum across all frames (most conservative = safest full coverage)
TEXT_ZONE_H=9999
for f in "$WORK_DIR"/probe_*.png; do
  OFFSET=$(magick "$f" -fuzz 8% -trim -format "%O" info: 2>/dev/null \
    | grep -oE '\+[0-9]+$' | tr -d '+')
  [ -n "$OFFSET" ] && [ "$OFFSET" -lt "$TEXT_ZONE_H" ] && TEXT_ZONE_H=$OFFSET
done

# Sanity check — if detection failed or video fills full frame, warn and stop
if [ "$TEXT_ZONE_H" -eq 9999 ] || [ "$TEXT_ZONE_H" -lt 50 ]; then
  echo "ERROR: Could not detect text zone (TEXT_ZONE_H=$TEXT_ZONE_H). Inspect probe frames manually." >&2
  exit 1
fi

# The trim offset is where the letterbox ENDS (below the caption text).
# The original caption text typically ends ~150px before this point.
# Add 150px buffer so the overlay fully covers the caption including descenders.
TEXT_ZONE_H=$(( TEXT_ZONE_H + 150 ))
echo "TEXT_ZONE_H=$TEXT_ZONE_H (with 150px buffer)"

Read $WORK_DIR/probe_1.png to visually confirm the text zone and identify:

• HOOK_BG — background color of the text zone (e.g. black or white) — match exactly
• HOOK_FG — foreground/text color (e.g. white or black) — match exactly
• The original hook text verbatim — read every line carefully, this is the input for copy generation

Step 3 — Generate Hook Copy Options

Using the original hook text and meme context, generate 3 hook options for the user to choose from.

Rules for each option:

• Keep the meme's structural premise (setup / punchline or before / after)
• Replace the original topic with a sobriety / Zero Proof angle
• The hook must feel native to the format — short, punchy, relatable to people in recovery or sober-curious
• Line 1: setup (max ~12 words). Line 2: punchline / contrast (max ~10 words, ends with : if the video reveals the punchline visually)
• Goal: stop the scroll → create curiosity → prime the viewer to care about sobriety

Example angles (use variety, don't repeat the same angle across 3 options):

• Identity shift: "Me before sobriety vs. me now:"
• Recovery pride: "My old self needed 5 drinks for this. / Me now, completely sober:"
• Relatability: "Nobody told me getting sober would make me THIS fun:"
• Social proof angle: "Everyone said I'd be boring without alcohol. / 365 days sober later:"
• Milestone: "1 year sober and I still don't miss it:"

Step 4 — Generate CTA Options

Generate 3 CTA options (2–3 lines each). Each should:

• Drive a specific action: download the Zero Proof app
• Feel conversational, not ad-copy stiff
• Create urgency or relevance to what the user just watched
• Vary the angle: curiosity, social proof, benefit, challenge

Examples:

• "Ready to track your own journey? / Download the Zero Proof app — / it's free to start."
• "See how far you've come. / Zero Proof helps you visualize / every day of your sobriety."
• "Your sober streak deserves a home. / Download Zero Proof today."

Step 5 — Propose Output Path

Generate:

• FOLDER — {MM-DD-YY}-{topic-slug} (today's date + 2–3 word slug from meme topic)
• FILENAME — 3–4 keyword-rich SEO words, no generics. Think: what would someone search on Instagram?
  • Good: sober-dance-freedom.mp4, alcohol-free-confidence.mp4
  • Bad: reel.mp4, video1.mp4, final.mp4

Step 6 — Confirmation Checkpoint (ONE PAUSE)

Present to the user in this exact format — do not render until confirmed:

PROPOSED REMIX — Zero Proof Reel

━━━ HOOK OPTIONS ━━━
[1] Line 1 / Line 2
[2] Line 1 / Line 2
[3] Line 1 / Line 2

━━━ CTA OPTIONS ━━━
[A] Full CTA text (2-3 lines)
[B] Full CTA text (2-3 lines)
[C] Full CTA text (2-3 lines)

━━━ OUTPUT ━━━
video-output/{FOLDER}/{FILENAME}

CTA duration: 3 seconds
Reply with hook # and CTA letter (e.g. "2A") or request changes.

Wait for the user's reply. Accept changes if requested, then confirm once more if the changes are substantial.

Step 7 — Generate Hook Overlay PNG

Use the chosen hook text. drawtext is unavailable — always use ImageMagick PNG overlay.

Resolve a font name that works with this ImageMagick build (use logical font names, NOT file paths):

# Use ImageMagick logical font names — file paths do NOT work with all builds
if [[ "$OSTYPE" == "darwin"* ]]; then
  FONT="Helvetica"   # always available on macOS via ImageMagick
else
  # Try common logical names; fallback to empty (ImageMagick default)
  magick -list font 2>/dev/null | grep -qi "DejaVu-Sans" && FONT="DejaVu-Sans" || FONT=""
fi
FONT_ARG=${FONT:+-font "$FONT"}

Split the hook into Line 1 and Line 2, then produce an overlay that fully covers the original text zone with the matching background color and text color:

Text is centered both horizontally and vertically within the detected text zone. Offsets are computed from TEXT_ZONE_H — no hardcoded values:

# Center lines within TEXT_ZONE_H (line spacing ~80px for pointsize 48)
LINE1_Y=$(( TEXT_ZONE_H / 2 - 40 ))
LINE2_Y=$(( TEXT_ZONE_H / 2 + 40 ))

magick -size ${VIDEO_W}x${TEXT_ZONE_H} xc:${HOOK_BG} \
  $FONT_ARG \
  -pointsize 48 \
  -fill ${HOOK_FG} \
  -draw "gravity North text 0,${LINE1_Y} '{{LINE_1}}'" \
  -draw "gravity North text 0,${LINE2_Y} '{{LINE_2}}'" \
  "$WORK_DIR/hook_overlay.png"

Read the PNG to verify text is centered within the zone and fully covers the original text before continuing.

Step 8 — Generate CTA Frame PNG

Use black background with white text (matches the style of the hook zone). Place each line with -draw text at explicit vertical offsets centered in the frame:

# LINE_SPACING ~80px for pointsize 54; adjust if more/fewer lines
magick -size ${VIDEO_W}x${VIDEO_H} xc:black \
  $FONT_ARG \
  -pointsize 54 \
  -fill white \
  -draw "gravity Center text 0,-50 '{{CTA_LINE_1}}'" \
  -draw "gravity Center text 0,50 '{{CTA_LINE_2}}'" \
  "$WORK_DIR/cta_frame.png"

For 3-line CTAs add a third -draw at text 0,150. Read the PNG to verify text is centered before continuing.

Step 9 — Extract Original Audio

ffmpeg -i "$WORK_DIR/original.mp4" \
  -vn -c:a copy \
  "$WORK_DIR/audio.aac" -y

Step 10 — Build Full Remixed Video (no audio)

Overlay the hook PNG on the full video — do not trim any content:

ffmpeg -i "$WORK_DIR/original.mp4" \
  -i "$WORK_DIR/hook_overlay.png" \
  -filter_complex "[0:v][1:v]overlay=0:0" \
  -an -c:v libx264 -preset fast \
  "$WORK_DIR/remixed_full.mp4" -y

Step 11 — Build Silent CTA Clip

CTA_DUR=3
ffmpeg -loop 1 -i "$WORK_DIR/cta_frame.png" \
  -t ${CTA_DUR} -r 30 -an \
  -c:v libx264 -preset fast -pix_fmt yuv420p \
  "$WORK_DIR/cta_silent.mp4" -y

Step 12 — Concatenate + Mux Final Output

Full video plays completely, then CTA is appended. Audio is looped to cover the full duration (original + CTA), so the reel sounds continuous throughout.

cat > "$WORK_DIR/concat.txt" << EOF
file '$WORK_DIR/remixed_full.mp4'
file '$WORK_DIR/cta_silent.mp4'
EOF

ffmpeg -f concat -safe 0 -i "$WORK_DIR/concat.txt" \
  -c:v libx264 -preset fast \
  "$WORK_DIR/video_only.mp4" -y

# Loop the original audio to cover the full output duration (video + CTA)
TOTAL_DUR=$(ffprobe -v quiet -show_entries format=duration -of csv=p=0 "$WORK_DIR/video_only.mp4")
ffmpeg -stream_loop -1 -i "$WORK_DIR/audio.aac" \
  -t "$TOTAL_DUR" \
  -c:a aac \
  "$WORK_DIR/audio_looped.m4a" -y

OUTPUT_DIR="$(pwd)/video-output/${FOLDER}"
mkdir -p "$OUTPUT_DIR"

ffmpeg -i "$WORK_DIR/video_only.mp4" \
  -i "$WORK_DIR/audio_looped.m4a" \
  -c:v copy -c:a copy -shortest \
  "$OUTPUT_DIR/${FILENAME}" -y

Verify:

ffprobe -v quiet \
  -show_entries format=duration,size \
  -show_entries stream=codec_name,width,height,duration \
  -of default "$OUTPUT_DIR/${FILENAME}"

Confirm: duration ≈ AUDIO_DUR + 3s (CTA appended), both video and audio streams match, h264 + aac.

Report the final output path to the user.

Key Principles

• Audio covers the full output — loop the original audio to fill video + CTA duration so the reel is never silent
• PNG overlay only — ImageMagick overlays work regardless of FFmpeg build (drawtext requires libfreetype which may be absent)
• Intermediate files are silent — no audio until the final mux step
• One confirmation only — do not ask the user anything else unless there is an actual error
• Working directory — all intermediate files go into $(pwd)/reel-remix-work/; never use /tmp
• Output path — $(pwd)/video-output/{MM-DD-YY}-{topic}/{3-4-word-seo-keyword-filename}.mp4