youtube-video-inspector

YouTube Video Inspector & Recreation Skill

This skill defines the standard procedure for inspecting, extracting, and visually analyzing YouTube videos when given a URL (with or without timestamps) to recreate typography, animations, transitions, or visual styles.

---

1. Core Rule: Never Claim Inability to View YouTube Videos

When the user provides a YouTube URL (e.g. https://www.youtube.com/watch?v=... or https://youtu.be/...):

• NEVER reply that you cannot view YouTube videos.
• ALWAYS use the headless extraction protocol below to download the relevant video format, extract video frames around the requested timecode, and visually inspect the frames via view_file.

---

2. Fast Headless Video Extraction Protocol

Step A: Ensure yt-dlp is Available

If yt-dlp is not installed, install it in Python environment:

python3 -m pip install yt-dlp

Step B: Download Small Single-Stream Video (No FFmpeg Merging Required)

To download in $< 3$ seconds without requiring external merging tools:

• Request format 134 (640x360 MP4 video-only) or format 18 (360p pre-muxed MP4):

python3 -m yt_dlp -f 134 "<YOUTUBE_URL>" -o "/Users/eram/.gemini/antigravity/scratch/ref_video_360p.mp4"

(If format 134 is unavailable, list formats via python3 -m yt_dlp -F "<URL>" and pick the smallest direct mp4 format 160, 133, 134, or 18).

---

3. Frame Sampling & Visual Analysis

Step A: Extract Frames Around Timestamp

Given timestamp $t$ (in seconds, e.g. t = 201s):

import cv2, numpy as np

video_path = '/Users/eram/.gemini/antigravity/scratch/ref_video_360p.mp4'
cap = cv2.VideoCapture(video_path)
fps = cap.get(cv2.CAP_PROP_FPS) or 30.0

target_sec = 201.0 # From URL &t=201s or user prompt
offsets = [-2.0, -1.0, 0.0, 1.0, 2.0, 3.0]

thumbs = []
for offset in offsets:
    f_idx = int((target_sec + offset) * fps)
    cap.set(cv2.CAP_PROP_POS_FRAMES, f_idx)
    ret, frame = cap.read()
    if ret:
        thumbs.append(cv2.resize(frame, (640, 360)))
cap.release()

# Build 6-frame comparison grid
row1 = np.hstack([thumbs[0], thumbs[1], thumbs[2]])
row2 = np.hstack([thumbs[3], thumbs[4], thumbs[5]])
grid = np.vstack([row1, row2])
cv2.imwrite('/Users/eram/.gemini/antigravity/scratch/youtube_ref_grid.jpg', grid)