Video Multiplier
One upload in, a text package out: chapters, five titles, a description and up to six post drafts.
Writes text, not clips: chapters, titles, a description and post drafts, all from the transcript. It never cuts video, uploads or posts.
00:00 the 9-minute budget · 02:15 killing the tour · 05:48 the cost of a form field
“We cut onboarding to 9 minutes” · “Your product tour is costing you accounts”
Full breakdown of the 9-minute onboarding rebuild: what we deleted, what we measured, links + timestamps.
We deleted our product tour and activation went up. The aha moment has a timestamp: find yours.
1/ We deleted our product tour. Activation went up. Here is the whole 9-minute rebuild:
HOOK: Your onboarding is 40 minutes. Ours is 9. Same product category.
This week we rebuilt onboarding around one number, and the number is nine minutes…
You already did the hard part: you recorded the video. This agent pulls the transcript, writes the chapters, five title options and the description, then drafts up to six posts from the same material: an X thread, a LinkedIn post, a newsletter section and two or three short-form scripts. It writes text and never cuts clips. Publishing stays yours.
One paste into your agent.
Your first content package, minutes later.
Install superagnt's Video Multiplier skill for me. It's the video-multiplier plugin from the github.com/superagnt/plugins marketplace, plus the superagnt MCP server (on Claude it's a connector I add in my browser; elsewhere I sign in through my browser). The setup guide for each kind of agent is at https://www.superagnt.com/install?skill=video-multiplier» installs the video-multiplier skill (read every file first, or the raw SKILL.md) and enables data:youtube, database
One of our uploads in, the full package out
The video: How to build an AI Agent Reddit monitoring tool for leads & trends, a 14:41 tutorial from our founder's channel. Two calls on a superagnt test workspace on Nov 3, 2026: 2 credits and 33 seconds. The transcript came back whole, 408 caption lines.
“Run the Video Multiplier skill on https://www.youtube.com/watch?v=jh6naUczWZ8 and turn it into chapters, titles, a description and the posts for X, LinkedIn, shorts and my newsletter.”
- 01Read the upload's details2.9 s wall time→ call
data_youtube_videos{ "action": "details", "video_id": "jh6naUczWZ8", "response_format": "json" }← result{ "video_id": "jh6naUczWZ8", "title": "How to build an AI Agent Reddit monitoring tool for leads & trends", "channel_name": "Jaen", "views": "489", "duration": "881", "created_at": "2026-04-16T10:58:04-07:00", "description": "Stop manually scrolling through subreddits! In this video, I show you how to set up a professional Reddit monitoring sys…" }» Trimmed: the rest of the description, the thumbnail and category
- 02Pull the whole transcript29.9 s wall time→ call
data_youtube_videos{ "action": "subtitles", "video_id": "jh6naUczWZ8", "response_format": "raw" }← result{ "lang": "en", "is_available": true, "subtitles": [ { "id": 1, "start": 0.8, "duration": 3.56, "text": "What's going on everyone? Today I'm" }, { "id": 2, "start": 1.92, "duration": 4.52, "text": "going to be showing you how to build a" }, { "id": 3, "start": 4.36, "duration": 4.48, "text": "Reddit listening agent. Um in this" } ] }» Trimmed: 405 more caption lines of 408, through the end at 14:41
Claude (Opus 5.5) in Claude Code wrote the package below from those 408 lines, by the skill's steps 2 and 3. Each draft names the moment it is built on, so you can check it against the video.
Chapters, at the real topic turns:
- 00:00 What a Reddit listening agent does
- 00:55 What you need: an agent and a server
- 02:54 Give the agent Reddit data
- 04:27 The cron mistake that makes it expensive
- 05:20 Step 1: pull posts by subreddit and keyword
- 05:49 Step 2: a cheap model checks each post
- 06:36 Step 3: score each post
- 06:59 How the script calls the data and the cheap model
- 08:00 Clone the starter repo and point it at your product
- 09:08 Step 4: send the scored posts to Slack
- 09:53 Deploy it on a schedule
- 11:21 Why a script plus a cheap model costs less
- 12:35 Pick the Slack channel and set the cron job
- 13:21 Add the API key
- 14:04 What you get: new posts, a few times a day
| Title option | Patterns it uses | Best hit rate |
|---|---|---|
| Reddit Lead Monitoring with AI Agents, Clearly Explained | "clearly explained", names "agent" | 27% |
| 4 Steps to a Reddit Agent That Finds Leads While You Sleep | number list, names "agent" | 27% |
| Reddit Lead Monitoring for Beginners: Your Agent Sets It Up | beginner, names "agent" | 20% |
| How to Build an AI Agent That Monitors Reddit for Leads | names "agent" | 16% |
| Why Your AI Agent Shouldn't Read Reddit Itself | names "agent" | 16% |
The five title options, ranked. Each is tagged with every title pattern it uses and the best hit rate among them, from our study of 1,412 long-form videos (below). The buckets overlap, so the rate breaks ties; it isn't a forecast.
The description:
Stop scrolling subreddits by hand. In this tutorial I build a Reddit listening agent: a script pulls new posts from the subreddits and keywords you pick, a cheap model scores each one, and the posts land in Slack with their scores a few times a day. Chapters as above, then the starter repo and superagnt for the Reddit data.
The X thread (built on the sequence from 04:34 to 12:16):
- 1/ Most Reddit monitoring agents burn money the same way: the main agent runs on a cron and makes every request itself, on its most expensive model. Here's the cheaper setup from my build video. (04:34)
- 2/ Step 1: a plain script pulls new posts for the subreddits and keywords you pick. (05:20)
- 3/ Step 2: a cheap model reads each post. "You really don't need an expensive model to kind of identify the post." (04:56)
- 4/ Step 3: score each post for relevance. Step 4: send the posts with their scores to Slack, or to email or a Google Sheet. (06:36)
- 5/ Then put it on a server with a cron job, so it runs a few times a day while your laptop is closed. (11:05)
- 6/ The cost: about 7 cents to run, with GPT-4o mini doing the reading. (12:16)
The LinkedIn post:
We wanted to find Reddit posts from people talking about churn, because those are the people who might want our product. So I built a Reddit listening agent. The mistake I see most: the main agent runs on a cron and makes every request itself, on its most expensive model (04:34). Mine splits the work: a script pulls the posts, a cheap model checks and scores each one, and the posts land in Slack with their scores. It runs a few times a day on a server, so it keeps going with my laptop closed. The full build is in the video.
Three short-form scripts, each built on one moment:
- Hook: "If your AI agent reads Reddit on a cron, you're overpaying." Beats: the main agent on its priciest model makes every request (0 to 12 s); split it, a script pulls posts and a cheap model scores them (12 to 25 s). CTA: the full build is on the channel. Built on 04:34.
- Hook: "Your AI agent stops working when you close your laptop." Beats: the copy on your laptop is only online while it's open; a copy on a small server runs 24/7, and ours runs a few agents for about $12 a month. CTA: how I set it up. Built on 01:53 and 02:30.
- Hook: "You don't need your smartest model to read Reddit." Beats: new posts on churn come in, a cheap model checks each one and scores it, and the posts land in Slack with their scores. CTA: watch the build. Built on 05:49.
The newsletter section:
Build of the week: a Reddit lead listener. A script pulls new posts from the subreddits and keywords you choose, a cheap model scores each one, and the posts land in Slack with their scores a few times a day. The trick is keeping your main agent out of the loop: "you really don't need an expensive model to kind of identify the post" (04:56). Watch the full build (14:41).
What each tool makes from your video
Most tools that repurpose a YouTube video cut it into clips. Video Multiplier writes the text around the video and the posts that point to it. Each row below was read on the tool's own home page on Nov 3, 2026.
| Tool | What it makes | Written posts | Publishes for you | Free to start |
|---|---|---|---|---|
| superagnt Video Multiplier | Chapters, five titles, a description, an X thread, a LinkedIn post, a newsletter section and short-form scripts, from the transcript | Yes, every piece is text | No, publishing stays yours | Free skill; this run cost 2 credits |
| OpusClip | Short clips from a long video, captions, reframing for each platform | Not described | Yes, "to all social platforms in one click" | Yes, 60 credits a month after a 7-day trial |
| Vizard | Short clips, vertical reframing, captions in 100+ languages, social captions and hashtags | Captions and hashtags only | Yes, "publish directly" | Yes, "Try it free" |
| CapCut | A video editor with short-video templates and automatic captions | Not described | Not described; it exports files for YouTube, TikTok and Reels | Yes, "Try online for free" |
| DeepReel | AI avatar and faceless explainer videos | Not described | Not described | No free plan listed; from $5 a month |
Which title patterns beat a channel's median
From our study of 1,412 long-form videos on 15 AI and builder channels, pulled Sep 30, 2026. A hit is a video with 3 times its own channel's median views; 13% of all the videos were hits.
| Pattern in the title | Videos | Median vs. channel | Hit 3x or more |
|---|---|---|---|
| "full course" or "(N hours)" | 38 | 2.8x | 45% |
| "clearly explained" or "explained" | 33 | 1.6x | 27% |
| A number list ("5 ways", "13 tools") | 59 | 1.3x | 27% |
| "learn X%" or "in N minutes" | 117 | 1.2x | 25% |
| Beginner, no code or 101 | 65 | 1.2x | 20% |
| Names "agent" | 196 | 1.2x | 16% |
| Names Claude | 278 | 1.2x | 14% |
| A money claim | 274 | 1.0x | 12% |
| "I" or "my" | 348 | 0.94x | 11% |
| A question mark | 55 | 0.74x | 2% |
»Every video scored as views divided by its own channel's median, which strips out channel size. Buckets overlap, and several are small (33 to 65 videos). Videos under about a week old were left out.
Three steps to your first content package.
- 1
Drop the link
Paste the upload, or point it at your channel so new videos queue themselves. It pulls the transcript and the stats and gets to work.
data_youtube_video_details · data_youtube_video_subtitles · agnt_db_insert
drop the linkexampleyoutu.be/xK4…→42 min · transcript pulledor point it at the channel · new uploads queue themselves
- 2
It packages the upload
Chapters with timestamps, five title options, the full description with links. Ready to paste into the studio the moment you open it.
data_youtube_video_subtitles · agnt_db_upsert
the studio packageexample00:00 the claim everyone gets wrong
04:12 the setup, start to finish
19:05 the benchmark run
chapters ×8 · 5 title options · description
- 3
Up to six posts, drafted
From the same transcript it drafts an X thread, a LinkedIn post, a newsletter section and two or three short-form scripts, each built on the video’s actual lines. You edit, you post.
agnt_db_select · agnt_db_upsert
the posts, draftedexamplethread · built on 04:12-11:40
short script ×2 · one moment each
newsletter section · the benchmark
you edit · you post · nothing auto-publishes
»After the first good run, the skill offers the always-on version: a small hosted runner on a schedule, so the work keeps happening with your laptop closed. You confirm the cadence before anything is scheduled.
Read every file before you install it.
SKILL.md tells your agent to read the video's details and transcript, stop and say so when there are no captions, write chapters at the real topic turns, five honest titles and the description, then draft the posts for the channels you named, each carrying its source timestamp. It never uploads, posts or edits your YouTube studio. Its first two headings are setup for the agent.
Plain markdown your agent reads while it runs. SKILL.md is the procedure. This is the plugin exactly as it installs from github.com/superagnt/plugins: 4 files, 8.8 KB.
Common questions
01How do I repurpose YouTube videos?
Start from the transcript, not the video file. Pull the captions, find the topic turns for chapters, pick the strongest sequence for a thread, and anchor each short script on one moment. Video Multiplier does those steps in your AI and hands you the drafts to edit.
02Can I repurpose a video that is already published on YouTube?
Yes. Paste the link: the skill reads the public video's details and its captions. The run on this page used a video published on Apr 16, 2026.
03Can you repurpose YouTube videos for TikTok, and how often should I?
This skill writes two or three short-form scripts per upload, each built on one moment, which you film for TikTok, Reels or Shorts. It doesn't cut the video itself. Post them as you film them; the skill can check your channel daily and queue each new upload.
04How accurate are AI YouTube transcripts?
The skill reads the captions YouTube has for the video, uploaded by the creator or generated by YouTube, and quotes them with timestamps so you can check any line. When a video has no captions, it says so and stops; it never packages from the title alone.
05Can I transcribe someone else's YouTube video?
The skill can read the captions of any public video, but it is built for your own uploads: it drafts posts in your voice from your words. Republishing someone else's words is a copyright question to settle before you post.
06How many pieces can I make from one video?
From the run on this page: chapters, five title options and a description for YouTube, then six posts: an X thread, a LinkedIn post, a newsletter section and three short-form scripts. The skill is free; the two calls cost 2 credits, and plans are $49, $149 or $399 a month.
Already connected? Your agent can list every packaged skill: agnt_tools_search + agnt_guidance_search