Skill Video Caption Overlay

Cleanup Review

python scripts/_cleanup.py is an optional cleanup helper for the Video Caption Overlay skill. It may recursively remove /tmp/.skill_state. Show the exact paths first, explain whether they are skill-private temporary paths or user state, and run the cleanup only after the user confirms.

Video Caption Overlay

Animated pill-style caption overlays for short-form video. No Premiere, no CapCut — pure Python.

Usage

uv run --with moviepy --with pillow scripts/overlay.py \
  --video base.mp4 \
  --output final.mp4 \
  --captions scripts/example_captions.json \
  --audio music.mp3 \
  --audio-start 8 \
  --audio-vol 0.5

No --audio if you want to keep the original video audio.

Custom fonts

--font-black /path/to/Montserrat-Black.ttf \
--font-bold  /path/to/Montserrat-Bold.ttf

Falls back to Montserrat from ~/.local/share/fonts/ if not specified.

captions.json format

Array of phases — each phase is a time window with one or more pill lines stacked vertically.

[
  {
    "start": 0,
    "end": 3.2,
    "y_frac": 0.06,
    "lines": [
      {
        "text": "POV:",
        "size": 28,
        "bold": true,
        "bg": [0, 195, 255],
        "fg": [0, 0, 0],
        "bg_opacity": 0.9,
        "px": 20, "py": 9, "r": 12
      },
      {
        "text": "drink more water",
        "size": 50,
        "bg": [255, 255, 255],
        "fg": [0, 0, 0]
      }
    ]
  }
]

PIL textbbox fix

PIL's textbbox((0,0), text, font) returns (x0, y0, x1, y1) where y0 is a non-zero offset (typically 7–15px depending on font size). Drawing text at (x, y) without compensating for this offset causes text to appear below the pill's visual center.

Fix implemented in pill():

bb    = draw.textbbox((0, 0), text, font=font)
x_off, y_off = bb[0], bb[1]
vis_w = bb[2] - bb[0]   # actual visual width
vis_h = bb[3] - bb[1]   # actual visual height

# Compensate offsets when drawing text
tx = cx - vis_w // 2 - x_off
ty = y - y_off
draw.text((tx, ty), text, font=font, fill=fg)

Emoji note

NotoColorEmoji.ttf fails with PIL at arbitrary sizes (bitmap font with limited supported sizes). Use text alternatives ("Free delivery" instead of "Free delivery 🚚") for reliable rendering.

Example output

See scripts/example_captions.json for the full 3-phase TikTok ad structure:

• Phase 1 (0–3.2s): Hook — top-screen pill stack
• Phase 2 (2.8–5.8s): Product claim — overlapping fade
• Phase 3 (5.3–8.0s): CTA — bottom-screen price + delivery + bio link