logo-popups

Logo popups: logos that pop out of your pointing finger

The user is usually not technical. Keep replies short, do the work for them, never ask them to run commands.

All paths are relative to this skill's folder (the folder this SKILL.md is in). Use its absolute path in commands. PY below means .venv/bin/python in this folder (.venv\Scripts\python.exe on Windows).

1. First run: set up (once)

Run bash scripts/setup.sh. It prints READY when done (1 to 3 minutes the first time).

• NEED_FFMPEG: install it (Mac: brew install ffmpeg, Windows: winget install ffmpeg), then run setup again.
• NEED_PYTHON: install Python 3.12 (Mac: brew install [email protected], Windows: winget install Python.Python.3.12), then run setup again.

2. Find the pointing and the words

Run both on their video:

• PY scripts/track.py <video>: tracks the hands, saves <video>.hands.json and prints pointing windows (times when a finger is held up).
• PY scripts/words.py <video>: prints when each word is said.

Match each window to the brand the speaker names in or near it. Whisper often mishears brand names: "cloud" is usually Claude, "codecs" Codex, "croc"/"grok" Grok, "chat GPT" ChatGPT, "Gemini", "cursor" Cursor. If the user already told you which logos, use those in the order they point. If no words name a brand and the user didn't say, ask one short question: which logos, in what order.

3. Write the plan

Save it next to the video as <video-name>_logos.json:

{"pops": [
  {"logo": "claude", "start": 1.12, "end": 1.85},
  {"logo": "codex",  "start": 2.10, "end": 3.05},
  {"logo": "grok",   "start": 3.45, "end": 4.50}
]}

• start: a beat after the hand settles (the window start), or on the word, whichever is later. end: when the finger drops or the next point starts. 0.7 to 1.2 s each looks best; never overlap two pops.
• logo: a preset name (below), any icon name from lobehub.com/icons (e.g. notion, figma, vercel; it tries <name>-color first), or a path to a PNG/SVG the user gave.

Options per pop (all optional):

key	what	default
label	text in the pill under the icon; "" hides it	brand name
bg	tile colour (hex)	preset, else white
fg	colour for one-colour SVG logos	white on dark tiles, black on light
accent	colour of the pop ring	preset, else logo colour
size	tile width as a share of the video width	0.24
logo_scale	logo size inside the tile	0.66
gap	how far past the fingertip the icon sits (in tile sizes)	0.85
nudge	extra [x, y] offset as a share of the width	[0, 0]
avoid_face	slide up if it would cover the face	true
burst	ring and sparks on pop	true

Presets: claude, anthropic, codex, openai, chatgpt, grok, gemini, cursor, perplexity, deepseek, copilot, mistral, meta.

4. Check, then render

1. Stills first (fast): PY scripts/render.py <plan.json> <video> <folder>/check.png --still 1.5,2.6,4.0, one time in the middle of each pop. Look at the PNG yourself. Fix anything that covers the face, gets cut off at the edge, or sits far from the finger (gap, nudge, size, timing).
2. Render: PY scripts/render.py <plan.json> <video> <video-name>_logos.mp4. Output: H.264 MP4, 1080 wide, same fps, original audio.
3. Tell them where the file is, and to watch it before posting.

Rules

• Never cover the face. Face avoidance is on by default; still check the stills.
• One logo per point. If they point twice at one brand, use two pops.
• Logos are trademarks of their owners; this is for commentary and reviews. Don't add logos the speaker doesn't mention unless the user asks.