A two-hour recording contains somewhere between fifteen and forty moments worth publishing. Finding them by hand takes longer than the recording did, which is why most long recordings get clipped once, badly, and then never touched again.
This guide covers the process that actually scales: how to get thirty publishable vertical clips out of one recording without watching it, and the three places the process reliably breaks.
Why manual clipping does not scale
Be precise about the cost, because it is worse than it feels.
Scrubbing a two-hour recording to find candidate moments takes roughly ninety minutes if you are quick and know what you are looking for. Cutting and exporting thirty clips takes another hour. Writing thirty titles and thirty sets of tags takes forty minutes. Call it three and a half hours per recording, before a single upload.
Now do that twice a week. Now do it for a roster of ten creators. The arithmetic stops working almost immediately, which is why agencies that try to scale manual clipping end up either capping their roster or quietly publishing far less than they claim to.
Step 1: scoring replaces scrubbing
The core idea is simple: instead of watching the video, let the AI analyse the whole of it and rate its moments.
Each moment gets one score from 1 to 10, for how strong the AI judges it.
Clip scoring explains what the score means and how to set the minimum for each creator.
You then set a minimum score. Every moment that reaches it becomes a clip; moments below it are never cut, so you never have to look at them. The minimum score is the single most important setting in the process, and the right value differs by creator, which is why it is set per creator rather than once for the whole account.
Step 2: from scored moments to clips
Two settings, both per creator, decide what comes out.
Length. Every clip is cut between the shortest and the longest clip length you set, 10 to 14 seconds by default.
Minimum score. Only moments that reach it become clips. Raise it for fewer, stronger clips; lower it for more from the same recording.
The output of this step is a set of clips, each with its score. This is the point where a human look actually makes sense: thirty clips with their scores, looked over in a few minutes, rather than a two-hour timeline. In WizeClip there is nothing to approve: only the few clips it flags, or one without a usable title, wait under To check, and Don't post keeps back any clip you do not want.
Step 3: cutting to vertical
Source recordings are usually horizontal. Every destination in the stack wants 9:16.
The naive approach, centre-cropping, fails often, because the subject is rarely centred for the whole segment. What you want is a crop that follows the subject, and framing that keeps important content away from the edges. WizeClip's portrait crop is a centre crop, so frame horizontal shows with the subject near the middle, or film vertically; either way keep important content away from the edges, since every feed platform overlays interface controls on top of the video.
Letterboxing is the one thing to avoid outright. A vertical clip with black bars looks reposted on every feed, and viewers swipe past it. Vertical clip specs per platform has the exact specs per destination.
Step 4: metadata, which is thirty times the work people expect
Thirty clips need thirty titles and thirty tag sets. That sounds like a small task and is not, because the destinations do not agree on what metadata even is.
One platform takes free-text tags typed one at a time. Another has its own fixed list of hashtags to choose from. A third has no tag field, so three hashtags go inline in the description. A fourth has its own list of categories, and a fifth has two separate taxonomies.
Writing one title and one tag set per clip and pasting it everywhere produces bad results on at least six of the ten destinations. Generating metadata per clip per destination is the only version that works, and it is the part of the workload that grows fastest. Writing titles and tags covers what each format wants.
Where the process breaks
Three failure modes, in order of how often they happen.
Minimum score set too low. You get sixty clips instead of thirty, half of them weak, and publishing them drags down your channel averages on every platform that tracks completion rate. More clips is not better. Raise the minimum score until the number of clips feels slightly too small.
Segments too long. The single most common cause of clips that get views but no completions. If you are torn between twenty-two seconds and forty, take twenty-two. This is covered in more depth in why your promo clips get zero views.
Publishing all thirty at once. Thirty clips from one account in one hour looks exactly like automation, because it is. Spread them across days. The feeds reward cadence over volume, and the video platforms do not care when you post, so there is no cost to spreading and a real cost to not.
A realistic output from one long recording
- Roughly 30 clips that reach the minimum score, from a long recording with steady action
- Each with its score out of 10, so you can see which came from the strongest moments
- Cut to vertical at the length you set, 10 to 14 seconds by default
- Metadata generated per clip, per destination format
- Published across ten destinations over five to seven days
Posted everywhere, which is the default, that is around three hundred individual uploads from one recording; set each clip to go to one platform only and it is thirty, with no platform seeing the same clip twice. Doing it by hand is not a workflow, it is a second job, which is the practical distinction between a clip tool and a distribution pipeline that the comparison pages work through.
