Video Repurposing Workflow: One Long Video Into 15 Assets

What this workflow does

You already make long video — webinars, podcast episodes, product walkthroughs, conference talks. This workflow turns one 30–60 minute video into a full derivative package: 5–8 vertical clips for Shorts/Reels/TikTok, 3–5 LinkedIn text posts, a quote-graphics set, a blog article, and an email — in roughly two hours of hands-on time instead of the two days it takes manually.

The key insight: the transcript is the master asset. Every derivative flows from a good transcript plus a model that has read the whole thing, not from scrubbing through video.

Prerequisites

• A source video (30+ minutes, decent audio)
• A transcription step: your clip tool's built-in transcription, or Whisper/Descript for standalone
• An AI clipping tool (OpusClip, Vizard, Descript, or Klap) — these find moments, reframe to vertical, and add captions
• An LLM for the text derivatives (Claude, ChatGPT)
• Your scheduling tool of choice

The workflow, step by step

Step 1: Transcribe and mark the map (15 minutes)

Upload the video to your clip tool (transcription happens automatically) or transcribe separately. Then have an LLM build a content map:

Here's a transcript of a [LENGTH] video: [TOPIC], speaker(s): [WHO].
Build a content map:
1. The 5-10 strongest standalone moments (timestamp + one-line summary
   + why it works as a clip: contrarian, tactical, story, or stat)
2. The 3 core arguments of the whole video
3. Every specific number, example, or quotable line

This map is your shot list for everything downstream. Skim it against your memory of the video — models occasionally rate a throat-clearing moment as a highlight.

Step 2: Cut the vertical clips (30 minutes)

Run the AI clipping tool, but don't accept its auto-selections blindly — cross-reference against your content map and keep the overlap plus anything the map rated highly that the tool missed (you can cut those manually by timestamp).

For each keeper: trim the first 1–2 seconds so the clip opens mid-thought (dead air kills retention), check the caption accuracy (names and product terms are the usual casualties), and replace the tool's auto-generated hook text with one derived from the clip's actual claim.

Target: 5–8 clips, 20–60 seconds each. Fewer, better clips beat a spray of mediocre ones.

Step 3: Write the text derivatives (30 minutes)

From the transcript and content map, generate each format with a format-specific prompt — not one "make me social posts" blast:

Using core argument #2 and its supporting examples from the transcript,
write a LinkedIn post. Structure: hook line that states the surprising
claim, 3-4 short paragraphs developing it with the specific example,
one-line takeaway. No hashtag spam, no "I'm excited to share."
Voice: [reference].

Repeat per argument and per platform. For the blog article, prompt for a restructure, not a transcript cleanup: "Reorganize this into an article with its own logic — the spoken order is not the written order." Edit everything; these are drafts.

Step 4: Build the quote graphics (15 minutes)

Take the 3–4 best quotable lines from the map into your template (Canva bulk-create from a spreadsheet works well, or an AI design tool). Verify quotes against the transcript verbatim — paraphrased "quotes" attributed to a real speaker are an embarrassment waiting to happen.

Step 5: Schedule as a campaign, not a dump (20 minutes)

Spread the package over 2–3 weeks: clips every 2–3 days, text posts interleaved on different days than clips, article published week one and linked from later posts, email in week two. Stagger platforms so the same asset doesn't hit every channel the same hour. Log each asset with its source video and format in a simple sheet.

Failure modes and fixes

• Clips get no retention. Usually the open is too slow or the clip needs context the viewer doesn't have. Cut harder into the moment, and prefer self-contained moments (a complete story or claim) over mid-argument excerpts.
• Everything sounds like a transcript. You skipped editing Step 3, or prompted for "summaries" instead of native-format writing. Written derivatives must be rewritten for the format, not extracted.
• Caption errors on names/jargon. Feed the tool a custom vocabulary list (most support it) with product names, speaker names, and industry terms before transcription.
• The package performs worse than the original. Check that each derivative carries one idea. The most common mistake is clips that try to compress the whole video instead of delivering one moment.

Turning it into a loop

After each package has run for 30 days, log per-asset performance and ask the model:

Here are 4 repurposing packages with per-asset results. Which moment
types (contrarian / tactical / story / stat), formats, and platforms
performed best? What should change in the next content map prompt
and clip selection criteria?

Two compounding effects follow. First, your selection criteria get sharper — you learn, with evidence, that (say) tactical moments outperform stories 3:1 for your audience on LinkedIn but reverse on TikTok. Second, the learnings flow upstream: you start recording long-form video differently — pausing before key claims, telling self-contained stories, stating numbers explicitly — because you now know what clips well. The loop doesn't just improve the derivatives; it improves the source.