Social Trailer and Hook Cutdown

Social Trailer and Hook Cutdown

What it does

Extracts hook moments from long-form video and produces platform-ready short clips (9:16, trailer teasers, opening hooks).

When to use

• Long video → TikTok/Reels/Shorts cutdown
• Creating trailer or teaser clips
• Optimizing first 3 seconds for retention

Inputs

• Long-form source video
• Target platform (9:16, duration cap)
• Optional transcript for hook detection

Outputs

• Hook segment identification (timestamp + rationale)
• Cutdown clip list with in/out points
• Optional 9:16 crop/reframe plan

Steps

1. Transcribe or analyze source for high-energy/key moments.
2. Score segments for hook potential (first 3s impact).
3. Select cutdown windows respecting platform duration limits.
4. Plan 9:16 crop/reframe if needed.
5. Export clip list for LosslessCut/ffmpeg.

Quality checklist

• Hook lands within first 3 seconds
  
• Clip duration within platform limit
  
• 9:16 framing preserves subject
  
• No mid-sentence cuts at start/end
  

What we found running it

Six tasks, each run with the skill installed and again with skills disabled, same task text. Both arms produced a defensible cut plan every time, and on the creative calls they often landed in the same place — both independently chose to drop the same two flat stretches from the founder interview in task 1, and both restructured it so the strongest line opens rather than arrives in sequence.

The reliable difference is that the skill arm hands you something you can run. Five of its six runs shipped an executable cut script — ffmpeg_cutdown.sh (1,511 B), cut_hook.sh (3,503 B), 04_ffmpeg_cutlist.sh (2,621 B), build_trailer.sh (2,289 B), ffmpeg_cut_commands.sh (1,357 B). The control arm shipped a script in none of its six. It stops at the EDL and the caption file, which is the right output if a human editor is doing the work and the wrong one if you wanted the rough cut itself.

The skill arm also tends to present ranked alternatives rather than one answer. Task 6 came back with three eight-second options ordered by priority, the third carried with an explicit warning that it reveals the outcome and is therefore contrast rather than suspense — an argument against its own third option.

Both arms were honest about the same structural limitation, which is worth recording because it applies to everything on this page: a transcript gives you words and rough timestamps, not frames. The skill arm marked two segments ESTIMATE in task 2 and noted in task 4 that the reversal shot has to be eyeballed against the source before anyone trusts the in-point; the control arm wrote a standing limitations section saying the same thing. Neither overclaimed. In task 6, where the 8-second cap was a hard placement limit, the control arm's margin reasoning was arguably the sharper of the two — it landed at 7.3 s and said the remaining 0.7 s is there to absorb export drift.

Cost, in task order: $0.35 / $0.50 / $0.41 / $0.39 / $0.43 / $0.41 with the skill, against $0.34 / $0.31 / $0.26 / $0.32 / $0.21 / $0.32 without.

What we did not test: none of the six cuts was rendered. Every task supplied a timestamped transcript and no video, so not one in-point in either arm was ever checked against a frame — including the ones inside the skill arm's ffmpeg scripts, which are executable but unverified. We also have no evidence about performance: nothing was published, so "hook" here means a structural argument, not a retention number.

Reference repos

• ViralCutter, NarratoAI, LosslessCut