Video Frame-by-Frame Hook and Scene Analysis

Pulls real frames, scene cuts, and word timings from a video to give a timestamped breakdown of what was on screen and what was said.

Creator / Videovideo-analysishook-analysis
by AgentVoltv1.0.0
Free to sign up · every skill included with AgentVolt Pro

Before & after

Same prompt, same agent — the only difference is the skill.

Prompt“Analyze the opening hook of this YouTube video for engagement”
Beforegeneric AI agent
I can help analyze video hooks. The most common approach is reading the transcript to see what's said in the first 5 seconds. If you paste the transcript or a transcript link, I can look for hook patterns—questions, curiosity gaps, that kind of thing. That said, the visual story matters too, but I'd be working mostly from your description of what's on screen.
Afterwith Video Frame-by-Frame Hook and Scene Analysis
VIDEO HOOK ANALYSIS: "Why Startups Fail" (2:43-3:18)

FRAME-BY-FRAME BREAKDOWN
0:00-0:08: Host looks directly at camera, close-up (forehead to shoulders)
  Verbal: "Nobody tells you this."
  Effect: Eye contact kills scroll (attention peak)
  Retention: 94% (internal benchmark)

0:08-0:16: Cut to B-roll of startup closing down (signage coming off wall)
  Verbal: "90% of startups fail by year three."
  Effect: Pattern-interrupt (visual shift), stakes raised
  Retention: 87%

0:16-0:24: Return to host, now leaning forward
  Verbal: "But nobody teaches WHY."
  Effect: Intrigue (promise of secret knowledge)
  Retention: 91%

HOOK VERDICT
Score: 8.2/10 (top 8% of this channel)
Strength: Direct address + visual cut at 0:08 (pattern-interrupt)
Weakness: Title oversells ("Why" answer delayed until 2:15)

RECOMMENDATION
Move the root cause to 0:20 to avoid drop-off. Tested on 12 peers: 3-4% retention lift.

About this skill

"Just read the transcript" misses what made the video work: the cut on the second word, the caption that contradicts the voiceover, the product shown before it's named. This skill gets Claude to actually look. It pulls frames at real scene cuts, lines each spoken word up with what was on screen at that moment, and returns a timestamped breakdown you can edit from.

What it does

  • Downloads public or local videos and pulls captions with word timings (or makes a transcript when there are none)
  • Finds hard cuts with ffmpeg scene detection, using thresholds tuned for edited content, talking heads, and screen recordings
  • Samples the opening 15 seconds densely and the rest at each cut, using contact sheets so the image budget lasts
  • Lines words up with frames so you can see what was on screen when each line landed
  • Measures pacing: time to first cut, average shot length, longest shots in the opening, and how often there's on-screen text
  • Labels each finding as observed, measured, or inferred, and lists which parts of the video it never viewed

What you get

A hook breakdown table (time, on screen, said, text overlay, note), a scene list with timestamps, pacing numbers, and a short list of things to copy or fix. Every row is tied to a frame that was actually viewed.

Best for

Short-form and YouTube creators studying competitor hooks, editors planning clips from long recordings, and teams that need a meeting or Loom summary that includes what was on the screen share.

SKILL.md preview

SKILL.md
---
name: video-frame-hook-scene-analysis
description: Watches a video by extracting real frames, detecting scene cuts with ffmpeg, and aligning word-level caption timings to what was on screen, then delivers a timestamped hook, scene, and pacing breakdown. Use when the user says "why does this video's hook work", "break down this TikTok frame by frame", "what's on screen in the first 3 seconds", "pull the scene changes from this video", "summarize this Zoom recording including the slides", "the transcript isn't enough, actually watch it", or "my videos lose people in the first few seconds, compare mine to this one".
---

# Video Frame-by-Frame Hook and Scene Analysis

… (sign up to view the full skill)
Sign up to view, copy, and install the full skill

More creator skills

View all Creator skills →