Vertical Clip Creator

Long video to shorts

by openfilm · MIT license · ★ 975 on GitHub

Turn a long video (a talk, podcast, interview, lesson, livestream or vlog) into short vertical clips that each stand alone and hold attention from the first second. Use when the person gives a long recording and asks for clips, shorts,…

---
name: long-to-short
description: Turn a long video (a talk, podcast, interview, lesson, livestream or vlog) into short vertical clips that each stand alone and hold attention from the first second. Use when the person gives a long recording and asks for clips, shorts, highlights or cut-downs for Douyin, Reels, Shorts or similar (长视频转短视频, 切片, 剪短视频, 高光, 拆条).
---

# Long video to shorts

Find the moments in this recording that people would stop scrolling for, and make each one a short a channel would post
as it is: it opens on its strongest real line, ends on its point, and makes sense to someone who never saw the long
video. The speaker, watching it, would say: yes, that is what I meant.

## What has to be true

- **The speaker's meaning, intact.** Nothing said in different places is joined into a claim they did not make; an
  "if", a "not" or an "in some cases" is never cut away; an answer keeps its question. A title or any other text on
  screen says only what the recording says, or is plainly yours.
- **It stands alone**: a viewer knows who is speaking and what they are answering.
- **Every word exact.** The transcript says exactly what was said; names, numbers and terms you cannot confirm are named

Overview

Find the moments in this recording that people would stop scrolling for, and make each one a short a channel would post as it is: it opens on its strongest real line, ends on its point, and makes sense to someone who never saw the long video. The speaker, watching it, would say: yes, that is what I meant.

Manual install (npx, ZIP)

Claude Codenpx skills add openfilm/openfilm --skill long-to-short -g -a claude-code

Claude Code (degit)npx degit openfilm/openfilm/packages/openfilm/skills/long-to-short#main ~/.claude/skills/long-to-short

Claude.ai / desktopZip the folder and upload it under Settings → Capabilities → Skills.

Codex, Cursor, other agentsPoint them at SKILL.md. The other files are plain text.

Connects with Agent Alley MCP · agentalley.io/mcp

Files included

SKILL.md2.4 KB

Read SKILL.md here

openfilm/main1 file shown

SKILL.md2.4 KB

Browse every file on GitHub↗

SKILL.md

Show the full text37 lines

long-to-short/SKILL.md37 lines · 2.4 KB

PreviewCode

Outline

RawView on GitHub

Long video to shorts

What has to be true

The speaker's meaning, intact. Nothing said in different places is joined into a claim they did not make; an "if", a "not" or an "in some cases" is never cut away; an answer keeps its question. A title or any other text on screen says only what the recording says, or is plainly yours.
It stands alone: a viewer knows who is speaking and what they are answering.
Every word exact. The transcript says exactly what was said; names, numbers and terms you cannot confirm are named in your reply. Each cut falls in a pause: no clip starts on a clipped word or stops mid-thought.
Made for a phone: the one who is speaking well framed at every moment, and nothing that matters where the app's buttons and captions sit.
Worth posting: each one holds up next to the best clips on the platform, as long as its moment needs and no longer. Watch each as a viewer would; a clip you would scroll past is not done.

What you have

The recording's transcript with every word's time (npx openfilm get asr, a WebVTT beside it), from which Studio draws the subtitles; the person sets how they look. A short is the same recording trimmed (#t=) and framed with CSS, never cut or re-encoded into new files; a page can also play the recording, silent, with its time set from t, when the picture has to move over it. Anything you can make titles and motion with, and your own built-in tools.

With the person

What they ask for comes before anything here. The recording, where the shorts will go, and how many they want (else the few best). Ask whether there are moments they must keep or must not use, without holding up the work. When done, list the shorts: for each, its length, its first line, and why you chose it.

1	---
2	name: long-to-short
3	description: Turn a long video (a talk, podcast, interview, lesson, livestream or vlog) into short vertical clips that each stand alone and hold attention from the first second. Use when the person gives a long recording and asks for clips, shorts, highlights or cut-downs for Douyin, Reels, Shorts or similar (长视频转短视频, 切片, 剪短视频, 高光, 拆条).
4	---
5
6	# Long video to shorts
7
8	Find the moments in this recording that people would stop scrolling for, and make each one a short a channel would post
9	as it is: it opens on its strongest real line, ends on its point, and makes sense to someone who never saw the long
10	video. The speaker, watching it, would say: yes, that is what I meant.
11
12	## What has to be true
13
14	- The speaker's meaning, intact. Nothing said in different places is joined into a claim they did not make; an
15	"if", a "not" or an "in some cases" is never cut away; an answer keeps its question. A title or any other text on
16	screen says only what the recording says, or is plainly yours.
17	- It stands alone: a viewer knows who is speaking and what they are answering.
18	- Every word exact. The transcript says exactly what was said; names, numbers and terms you cannot confirm are named
19	in your reply. Each cut falls in a pause: no clip starts on a clipped word or stops mid-thought.
20	- Made for a phone: the one who is speaking well framed at every moment, and nothing that matters where the app's
21	buttons and captions sit.
22	- Worth posting: each one holds up next to the best clips on the platform, as long as its moment needs and no
23	longer. Watch each as a viewer would; a clip you would scroll past is not done.
24
25	## What you have
26
27	The recording's transcript with every word's time (`npx openfilm get asr`, a WebVTT beside it), from which Studio draws
28	the subtitles; the person sets how they look. A short is the same recording trimmed (`#t=`) and framed with CSS, never
29	cut or re-encoded into new files; a page can also play the recording, silent, with its time set from t, when the
30	picture has to move over it. Anything you can make titles and motion with, and your own built-in tools.
31
32	## With the person
33
34	What they ask for comes before anything here. The recording, where the shorts will go, and how many they want (else the
35	few best). Ask whether there are moments they must keep or must not use, without holding up the work. When done, list
36	the shorts: for each, its length, its first line, and why you chose it.
37