YouTube video workflow · production operations · release checks
I turned a noisy clip collection into a repeatable video pipeline.
It finds clips, narrows the review set, builds drafts, checks the output, and leaves release approval to me.
Over a 32-day production window, 33,365 clips entered the funnel; the run recorded 85 published output records. The operation keeps moving when a source fails and stops a video when its final checks do not pass.
Production window: July 22–August 23, 2026.
The record
A noisy clip collection became a repeatable path to publish-ready output.
The system does the volume work and the owner reviews the narrowed lineup before making the release call.
33,365
clips harvested in 32 days
32
days in the production window
85
published output records
40–90 min
daily work window before repetition compounds
Why I built it
Find more. Watch less.
A gaming channel built from raw clips turns into a chore fast: thousands of files and no reliable way to tell which five minutes are worth watching back. I wanted a system that does the tedious narrowing automatically and leaves the judgment call, is this moment actually good, to me.
The two tiers are intentional: broad category search finds channels I would never know to look for, while the 117-creator watchlist keeps channels I trust visible. The design goal is not to remove taste. It is to give taste a much better starting point.
How it works
Discover → Collect → Narrow → Review → Approve.
85 published output recordsThis is a production-ledger total, not the next stage after the 53 approvals shown above. Review decisions and output records measure different populations.
What it actually took to build
Seven systems make the five-stage story work.
The macro diagram keeps the story easy to follow; this is the complete operating chain underneath it.
01
Finding the material. Three platforms, 16,479 channels surfaced by broad search, plus 117 creators I deliberately keep on the watchlist.
02
Cutting the noise. A scoring model over view velocity, clip density, streamer weighting, language, and dozens of title keywords.
03
Watching the video itself. Scene detection and action scoring pick the moment. A camera-box detector finds the streamer's face so the vertical crop keeps their reaction in frame.
04
Making it publishable. Compositing stitches the video layers together; the overlay package and intro and outro cards frame the result; sound levels are balanced automatically.
05
Naming it. Titles, descriptions, tags, and ranked thumbnail candidates, generated automatically.
06
Owner review. The shortlist lands in a spreadsheet. Approve, reject, pick a thumbnail.
07
Shipping it. Upload to YouTube and TikTok, scheduled to go out on its own once I have approved it.
Repetitive narrowing, not judgment.
The pipeline handles discovery, assembly, and routine checks. I still review the shortlist, approve clips, and choose the thumbnail.
One bad source cannot stop the queue.
Source errors are logged and isolated. A trim-duration check protects the approved lineup, and the release gate has survived 3,752 logged errors.
Three constraints shaped the system.
A limited daily work window, a noisy source pool, and a human approval gate determined what needed automation and what needed me.
Make every ledger’s scope explicit earlier.
Review decisions and production outputs should never be mistaken for the same population.
Why it earns a place
The pipeline turns repetitive discovery into an owner-approved production queue while keeping the final taste and release call human.
The audience result
The channel is young, and it started from a standing start.
The pipeline above is the production discipline. This is what the channel did with it in its first 28 days live. Growth this fast is normal for a brand-new channel before the curve settles; it is not a claim that this rate holds.
47.4K
views, first 28 days
304.7
watch-time hours, first 28 days
1,693
average views per day
+20
subscribers, first month
Self-reported from YouTube Studio, first 28 days the channel was live. Kept separate from the production receipt on purpose: that receipt proves the pipeline is real and selective; this shows what the channel did with it.
Evidence receipt
The receipt behind the production result.
Open the workflow receipt
The receipt supports the production result: how broad the pool was, what reached review, and what was approved or used. Audience figures are self-reported from YouTube Studio, dated separately below, and kept apart from the pipeline counts on purpose. The public page shows the checked summary; the raw working record remains private.
Read the scopes separately: 53 approvals, 97 hand-rejections, 150 reviewed, and 1,244 pending are decisions on 1,394 review-list entries. The 81 source clips marked used and 85 published output records come from separate production tables, so they are not a one-to-one conversion. The 33,284 remainder means clips not marked used in output records (33,365 minus 81); it includes items still pending review.
Operational effort: the ledger logged 3,752 source or processing errors over the snapshot. The pipeline isolates a failing source rather than treating every error as a failed run.
The claim
33,365 clips entered the funnel; 150 reached owner review and 85 became published output records.
The checks
Hand review, pending work, source failures, and the owner release decision.
The scale
Sources monitored, production days, languages, and the separate review and production populations.
117
creators explicitly on the watchlist
32
days in the production window
53
approved by hand
33,365
clips harvested
81
clips used
64.7%
hand rejection rate
1,244
pending review
1,394
queued for review
33,284
not marked used (includes pending)
99.76% not-used share of harvested clips
3
review languages
0.243%
selection rate
16,479
sources monitored
150
reviewed by the owner
97
rejected by hand
85
published output records
If this sounds useful
Build the review loop.
The same discipline applies well beyond video: narrow with rules, and keep the owner on the final call.