longform-retention
High-retention YouTube long-form editor (MrBeast-memo style structure, widely reported but leaked/unofficial): pays off the title and thumbnail in the first minute, cold-opens on the best moment, re-engages around 3:00 and 6:00 and then every ~3 minutes, stair-steps the stakes, k
Install
npx skills add https://github.com/Viprasol-Tech/claude-video-editing-skills/tree/main/plugins/video-studio/skills/longform-retention
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install viprasol-tech-claude-video-editing-skills@llmmart
git clone https://github.com/Viprasol-Tech/claude-video-editing-skills.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole viprasol-tech/claude-video-editing-skills collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Long-form Retention
A 8-20 minute YouTube video edited so the retention graph stays flat: promise early, escalate, end on the payoff.
- Best for: challenges, experiments, story videos, entertainment talking-head videos, list videos
- Inputs: A-roll (one or many clips), B-roll, the title and thumbnail idea (the promise the first minute must pay off)
- Default skin:
chapter-pro· alternativesbold-bounce,news-desk,essay-mode(calmer topics) - Formats: 16:9 1920x1080 master, chapters, thumbnails, 9:16 cutdowns
- Duration target: 8-20 min; genre preset
longform-retention
Before you start
- Needs video-studio-core in the same skills folder.
CORE=video-studio-core,SKILLS= the folder holding both. - First time:
python CORE/scripts/check_env.py. Preset:python CORE/scripts/editor.py genre longform-retention. - The
--genrepreset sets autocut's pause cut (0.45 s) and run.py respects it; pass--max-gaponly to override. - Ask for the title and thumbnail. The first 60 s must deliver that promise. Match intensity to the topic: the fast style can hurt calm educational content (use
video-essayortutorial-screenthere).
How a pro cuts this
| time | what happens |
|---|---|
| 0:00-0:15 | cold open: the title promise, shown not told; the best visual moment |
| 0:15-1:00 | stakes and rules; prove the thumbnail; no channel intro, no "subscribe" |
| 1:00-3:00 | stop telling, start showing; compress time; quick scene changes |
| ~3:00 | re-engagement: a spectacle, twist, reveal or new stake |
| 3:00-6:00 | attach the viewer to people; escalate |
| ~6:00, then every ~3 min | re-engagement beats; stakes stair-step up |
| back half | energy can dip but never sags; no "in conclusion" |
| payoff | the moment the title promised; end within 1-2 s after it, then end screen (5-20 s) |
- Pacing: ASL 3-6 s, 2-5 s in minute one. A visual change at least every 5 s (counter, zoom, B-roll, SFX, angle). Vary shot length.
- Cut rules: cut every "um", restart and dead moment; keep reactions that carry emotion. Time compression: show 3 seconds of a 30-minute task with a
timer/clock chip. - J/L cuts: reactions as L-cuts over the next action; J-cut the next scene's sound to pull forward.
- Transitions: cuts, zoom punches, whips with whooshes between scenes; one transition language.
- Music/SFX: music changes per section (tension under stakes, release after wins), 18-24 LU under speech; SFX 3-6/min (whoosh on moves, ding on counts, riser into reveals, silence 0.5 s before the payoff).
- Colour:
punchy/gaming-vivid0.6-0.7 for entertainment; skin natural. - Captions: sidecar SRT; on-screen text for numbers, stakes, rules, time ("DAY 3").
- Client judges: first-day AVD/AVP, CTR, the second-by-second retention graph, the first-minute drop.
- Common mistakes: channel intro before the hook, the promise delivered at minute 8, signalling the end early, flat stakes, uniform pacing, an outro after the payoff.
Make the edit
python CORE/scripts/run.py - --clips footage/*.mp4 --auto story --style SKILLS/chapter-pro --genre longform-retention \
--smart-cuts --retakes --broll broll/ --model small --work yt --stop-after plan
python CORE/scripts/story.py outline --work yt --genre longform-retention # hooks + beat map vs structure
python CORE/scripts/story.py reorder --work yt --order "s88-s90,s1-s87,s91-s140" --genre longform-retention --format 16:9
python CORE/scripts/run.py - --timeline yt/story_timeline.json --style SKILLS/chapter-pro --genre longform-retention \
--smart-cuts --work yt2 --stop-after plan
python CORE/scripts/editor.py review yt2/plan.json --genre longform-retention --fix # retention score, interrupts
python CORE/scripts/render.py yt2/plan.json --frames 5,60,180,360
python CORE/scripts/render.py yt2/plan.json
python CORE/scripts/finish.py yt2/plan.json --platform youtube
Read story_flags.md after a reorder; reorder only whole sentences for a cold open.
Refine the plan
- Re-engagement marks: list the times near 3:00, 6:00, 9:00... and put a reveal there:
titlecard("ROUND 2"), acounterof money/stakes, atimer, a new location chip, ariser->impactSFX pair (riserat hit - 1.2 s). - Stakes on screen:
{"type": "counter", "from": 0, "to": 10000, "prefix": "$", "label": "prize pool"};{"type": "timer", "from": 10, "label": "seconds left"};pollfor "who wins?". - Pattern interrupts:
speaker.zoomsalternating 1.12/1.07 where review.md reports stretches over 5 s. - Chapters: top-level
"chapters"(first at 0, >= 3, each >= 10 s), named as hooks ("The $10,000 mistake"). - End: delete anything after the payoff except 5-20 s of end-screen space; no "thanks for watching" over it.
- Sound:
"music": {"src": "act1.wav", "under": 20, "duck": "speech-aware", "fitTo": "beats"}, music premixed per section if it changes;"roomTone": true. - Look:
"color": {"look": "punchy", "intensity": 0.6}. - Face-aimed punch-ins: with
W/track.jsonthe zooms centre on the face instead of the frame centre: afterrun.py ... --stop-after cut(--stop-after timelinefor a--timelinespec) runpython CORE/scripts/track.py analyze W/cut.mp4 --work W, thenmake_plan.py(it copies the track into the plan). Per zoom,"cx"/"cy"(0-1) aim elsewhere;"toFace": falsekeeps centre.
Deliver
python CORE/scripts/deliver.py yt2/plan.json --to youtube --out-dir yt2/deliver
python CORE/scripts/qc.py yt2/deliver/final_youtube.mp4 --plan yt2/plan.json --platform youtube
- Look at
thumb_*.png; the thumbnail moment must appear in the first minute. - Cutdowns:
clip-miner-proonyt2/final.mp4for 3-5 Shorts.
Check before delivering
- Title/thumbnail promise visible in the first 60 s; nothing before the hook.
- Re-engagement beat near 3:00, 6:00 and every ~3 min after; stakes rise.
- No 5+ s stretch without a visual change (
review.md); pacing varies. - Ends within 2 s of the payoff;
qc.pypasses. - Tell the user: where each re-engagement is, chapter list, files.
Example requests
- "Edit my challenge video MrBeast-style - hook in the first 10 seconds, re-engage every 3 minutes."
- "My retention drops at 0:40, recut the first minute so it pays off the thumbnail."
- "Fast-paced 12-minute YouTube edit with counters, timers and SFX."
Files (claude-video-editing-skills)
-
SKILL.md 6.9 KB
--- name: longform-retention description: "High-retention YouTube long-form editor (MrBeast-memo style structure, widely reported but leaked/unofficial): pays off the title and thumbnail in the first minute, cold-opens on the best moment, re-engages around 3:00 and 6:00 and then every ~3 minutes, stair-steps the stakes, keeps constant visual change (zooms, counters, timers, SFX, captions) and ends abruptly after the payoff. Use when the user says: edit for retention, YouTube video editing, MrBeast style, challenge video, make my long video hold attention, fix my retention graph, fast-paced YouTube edit. Default skin chapter-pro; any Video Studio skin works." license: MIT --- # Long-form Retention A 8-20 minute YouTube video edited so the retention graph stays flat: promise early, escalate, end on the payoff. - **Best for:** challenges, experiments, story videos, entertainment talking-head videos, list videos - **Inputs:** A-roll (one or many clips), B-roll, the title and thumbnail idea (the promise the first minute must pay off) - **Default skin:** `chapter-pro` · alternatives `bold-bounce`, `news-desk`, `essay-mode` (calmer topics) - **Formats:** 16:9 1920x1080 master, chapters, thumbnails, 9:16 cutdowns - **Duration target:** 8-20 min; genre preset `longform-retention` ## Before you start - Needs **video-studio-core** in the same skills folder. `CORE` = `video-studio-core`, `SKILLS` = the folder holding both. - First time: `python CORE/scripts/check_env.py`. Preset: `python CORE/scripts/editor.py genre longform-retention`. - The `--genre` preset sets autocut's pause cut (0.45 s) and run.py respects it; pass `--max-gap` only to override. - Ask for the **title and thumbnail**. The first 60 s must deliver that promise. Match intensity to the topic: the fast style can hurt calm educational content (use `video-essay` or `tutorial-screen` there). ## How a pro cuts this | time | what happens | | --- | --- | | 0:00-0:15 | cold open: the title promise, shown not told; the best visual moment | | 0:15-1:00 | stakes and rules; prove the thumbnail; no channel intro, no "subscribe" | | 1:00-3:00 | stop telling, start showing; compress time; quick scene changes | | ~3:00 | re-engagement: a spectacle, twist, reveal or new stake | | 3:00-6:00 | attach the viewer to people; escalate | | ~6:00, then every ~3 min | re-engagement beats; stakes stair-step up | | back half | energy can dip but never sags; no "in conclusion" | | payoff | the moment the title promised; end within 1-2 s after it, then end screen (5-20 s) | - **Pacing:** ASL 3-6 s, 2-5 s in minute one. A visual change at least every 5 s (counter, zoom, B-roll, SFX, angle). Vary shot length. - **Cut rules:** cut every "um", restart and dead moment; keep reactions that carry emotion. Time compression: show 3 seconds of a 30-minute task with a `timer`/clock chip. - **J/L cuts:** reactions as L-cuts over the next action; J-cut the next scene's sound to pull forward. - **Transitions:** cuts, zoom punches, whips with whooshes between scenes; one transition language. - **Music/SFX:** music changes per section (tension under stakes, release after wins), 18-24 LU under speech; SFX 3-6/min (whoosh on moves, ding on counts, riser into reveals, silence 0.5 s before the payoff). - **Colour:** `punchy` / `gaming-vivid` 0.6-0.7 for entertainment; skin natural. - **Captions:** sidecar SRT; on-screen text for numbers, stakes, rules, time ("DAY 3"). - **Client judges:** first-day AVD/AVP, CTR, the second-by-second retention graph, the first-minute drop. - **Common mistakes:** channel intro before the hook, the promise delivered at minute 8, signalling the end early, flat stakes, uniform pacing, an outro after the payoff. ## Make the edit ```bash python CORE/scripts/run.py - --clips footage/*.mp4 --auto story --style SKILLS/chapter-pro --genre longform-retention \ --smart-cuts --retakes --broll broll/ --model small --work yt --stop-after plan python CORE/scripts/story.py outline --work yt --genre longform-retention # hooks + beat map vs structure python CORE/scripts/story.py reorder --work yt --order "s88-s90,s1-s87,s91-s140" --genre longform-retention --format 16:9 python CORE/scripts/run.py - --timeline yt/story_timeline.json --style SKILLS/chapter-pro --genre longform-retention \ --smart-cuts --work yt2 --stop-after plan python CORE/scripts/editor.py review yt2/plan.json --genre longform-retention --fix # retention score, interrupts python CORE/scripts/render.py yt2/plan.json --frames 5,60,180,360 python CORE/scripts/render.py yt2/plan.json python CORE/scripts/finish.py yt2/plan.json --platform youtube ``` Read `story_flags.md` after a reorder; reorder only whole sentences for a cold open. ## Refine the plan - **Re-engagement marks:** list the times near 3:00, 6:00, 9:00... and put a reveal there: `titlecard` ("ROUND 2"), a `counter` of money/stakes, a `timer`, a new location chip, a `riser` -> `impact` SFX pair (`riser` at hit - 1.2 s). - **Stakes on screen:** `{"type": "counter", "from": 0, "to": 10000, "prefix": "$", "label": "prize pool"}`; `{"type": "timer", "from": 10, "label": "seconds left"}`; `poll` for "who wins?". - **Pattern interrupts:** `speaker.zooms` alternating 1.12/1.07 where review.md reports stretches over 5 s. - **Chapters:** top-level `"chapters"` (first at 0, >= 3, each >= 10 s), named as hooks ("The $10,000 mistake"). - **End:** delete anything after the payoff except 5-20 s of end-screen space; no "thanks for watching" over it. - **Sound:** `"music": {"src": "act1.wav", "under": 20, "duck": "speech-aware", "fitTo": "beats"}`, music premixed per section if it changes; `"roomTone": true`. - **Look:** `"color": {"look": "punchy", "intensity": 0.6}`. - **Face-aimed punch-ins:** with `W/track.json` the zooms centre on the face instead of the frame centre: after `run.py ... --stop-after cut` (`--stop-after timeline` for a `--timeline` spec) run `python CORE/scripts/track.py analyze W/cut.mp4 --work W`, then `make_plan.py` (it copies the track into the plan). Per zoom, `"cx"`/`"cy"` (0-1) aim elsewhere; `"toFace": false` keeps centre. ## Deliver ```bash python CORE/scripts/deliver.py yt2/plan.json --to youtube --out-dir yt2/deliver python CORE/scripts/qc.py yt2/deliver/final_youtube.mp4 --plan yt2/plan.json --platform youtube ``` - Look at `thumb_*.png`; the thumbnail moment must appear in the first minute. - Cutdowns: `clip-miner-pro` on `yt2/final.mp4` for 3-5 Shorts. ## Check before delivering - Title/thumbnail promise visible in the first 60 s; nothing before the hook. - Re-engagement beat near 3:00, 6:00 and every ~3 min after; stakes rise. - No 5+ s stretch without a visual change (`review.md`); pacing varies. - Ends within 2 s of the payoff; `qc.py` passes. - Tell the user: where each re-engagement is, chapter list, files. ## Example requests - "Edit my challenge video MrBeast-style - hook in the first 10 seconds, re-engage every 3 minutes." - "My retention drops at 0:40, recut the first minute so it pays off the thumbnail." - "Fast-paced 12-minute YouTube edit with counters, timers and SFX."
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.