Repurposing long videos into short clips: the chop-and-publish workflow

A 60-minute interview should net you 4 to 6 clips worth posting. Not 40. The reason most channels stall at "we tried repurposing" is that someone fed an episode into Opus Clip, got 28 vertical clips back, posted half of them, and watched the analytics flatline.

The chop is not the workflow. The chop is step three.

1. Find the moments before you touch a tool

Before any AI sees your file, scrub the source and write down timestamps for three things:

• Peak moments — the line where the guest's voice changes, the laugh, the "wait, say that again."
• Q&A pairs — a tight question followed by a tight answer. These are the most reliably postable units in any interview.
• Story beats — anecdote, conflict, resolution. If it has a beginning and an end inside 90 seconds, it travels.

This is the part Justin Welsh has been preaching for years: pillar first, spokes second. He does it with newsletters, but the principle ports cleanly — one strong long-form piece becomes 6 to 12 derivatives, not "as many as the tool spits out" (Tweet Hunter on Welsh's system).

The strongest repurposing systems start from moments with tension, clarity, or surprise — not from a fixed quota of "three Reels and two Shorts" (Loonacast on podcast repurposing).

If you skip this step, you are outsourcing taste to a virality classifier. That goes badly.

2. The AI auto-clip tool landscape (May 2026)

The market has converged. Here is what is actually worth your money:

Opus Clip — $15/mo Starter, $29/mo Pro (300 input minutes). The Pro tier is the one with the editor; Starter only lets you download what the AI generated, no edits (OpusClip pricing; Ascynd breakdown). Strongest for reframe and B-roll automation.

Vizard — Free tier is 60 input minutes with watermark; Creator at $14.50/mo annual unlocks 4K and removes the watermark (Vizard pricing). The cheapest serious option.

Submagic — $19/mo Starter (15 videos, 2-min cap), $39/mo Pro (40 videos, 5-min cap), $69/mo Business with 4K and 30-min cap (Submagic pricing). Caption styling is the best in class; clip selection is weaker than Opus.

Munch — $38/mo Essential annual for 500 minutes (or $48/mo monthly). Pitches itself on trend-matching the clip topics to what is hot per platform (Munch pricing). Useful if your niche moves fast.

ClipAnything — Opus's natural-language selector ("find every moment where the guest disagrees"). Bundled into Opus Pro, not a separate product. Worth the upgrade if your source is unstructured.

Descript — different category. Not an auto-clipper. You edit by editing the transcript. $24/mo Creator. Pair it with one of the above for the polish pass; do not expect it to find moments for you.

3. The contrarian: AI auto-clip output is roughly 70% throwaway

The auto-clip tools sell a fantasy that you upload a podcast and walk away with 30 viral clips. In practice, on a typical interview, two thirds of the clips a tool generates are unusable — wrong context, the punchline cut off, a hook that misrepresents what was said, or a "viral score" of 87 attached to a clip that is genuinely boring.

This is not a bug in any one tool. It is what happens when a model trained on engagement signals tries to compress nuance.

So the workflow is not "review the AI's clips and post the good ones." It is "use the AI as a fast first pass, then throw most of it away." Plan for a 20–30% keep rate. Anything higher and you are not reviewing — you are rubber-stamping.

4. Manual vs AI — when each one wins

AI wins when the source is dense and structured: a webinar with chapter markers, a podcast with clear Q&A turns, a keynote with slide changes. The tool's transcript-aware clipping handles those cleanly.

Manual wins when the source is conversational, has long setups, or relies on visual context the model cannot see (a product demo, a screen share, a guest reacting to something off-camera). In those cases, opening Descript or your NLE and pulling clips by hand is faster than fixing what the auto-clipper got wrong.

A reasonable rule: if you can describe the clip in one sentence before watching the video, AI will find it. If you can only describe it after watching, do it manually.

5. The polish pass (this is where clips get won or lost)

Three things matter, in order:

Hook rewrite. The AI-generated title is almost always too literal. Rewrite the first 1.5 seconds — both the on-screen text and the spoken cold-open if you can re-edit. "Here's what nobody tells you about [thing]" is dead. Specificity wins: "We lost $40K on this in Q3."

Captions. Submagic and Opus both do solid auto-captions, but check three things on every clip: the speaker's name spelling, any product name, and any number. Those are the words people screenshot. They are also the words AI captions get wrong most often.

Vertical reframe. Auto-reframe is genuinely good now — both Opus's ReframeAnything and Vizard handle two-shot interviews well. Manually verify the cut on any clip where the speaker gestures or holds something up.

Skip color grading on clips. It does not move the needle on a 9:16 feed and it eats time you do not have.

6. Cadence — do not ship 50 mediocre clips a week

The most common mistake after adopting an auto-clipper: cranking the volume from 3 clips a week to 30, then watching the average view count drop by 80%.

The platforms reward consistency and a recognizable voice. They punish channels that suddenly look like a content farm. If your weekly posting jumps from 3 to 30, the algorithm reads "this account changed hands" and throttles distribution while it re-classifies you.

A defensible cadence for a weekly long-form show: 4 to 6 clips per episode, posted across 5 to 7 days, with the strongest clip going out on day 1 and the weakest never going out at all. That is roughly a Welsh-style spoke count and it matches what auto-clippers produce when you are honest about quality.

The takeaway

Set a hard cap before you start: this episode gets 5 clips, period. Then use the AI tool to generate 15–20 candidates, ruthlessly cut to the cap, and spend the saved time on hook rewrites instead of producing more clips nobody will watch.