FloristMark · Lesson production

From three cameras to every language

How one filmed floral-design lesson becomes a finished video in every language we publish — and exactly where the hired video editor's job starts and stops. Everything below the picture-lock gate is downstream of one file. Everything above it is how that file gets made.

Stages 0–1 built & live Stage 2 being hired for Stages 4–5 sketch only 5–7 lessons / week Updated 5 Aug 2026
The pipeline

Six stages, one gate, then a fan-out

Read down the spine. Colour tells you who owns the work, not what the work is — because every place the colour changes is a place a lesson can sit and wait.

Creator Organizer app The hired editor Abel AWS render Gate
STAGE 0

Capture On site · the creator

Three cameras roll on one lesson — Front View, Side View, Overhead — alongside a dedicated audio recorder. Four recordings, not three.

The creator pauses and resumes as she works, so any angle can arrive as several files numbered -1, -2. That is normal, and putting them back in order is part of the editor's job every single time.

Files land in Google Drive
STAGE 1

Organize Recording-organizer app · live

The lesson gets a code and its own Drive folder. Every file is previewed, tagged with the camera it came from, and renamed to a canonical name so it sorts predictably:

FM-0042__bridal-bouquet-mechanics__front-view.mp4

Gap found in the code · being fixed

Today the app marks a lesson “ready” once the three cameras are present. It never checks for the audio recording.

Since the dedicated recorder is the authoritative voice track, the app would happily hand the editor a lesson with no usable sound. A lesson now waits at “Ready — waiting on audio” until all four files exist.

Status reaches ready_for_production — enters the editor's queue
STAGE 2

Edit The hired editor · DaVinci Resolve Studio

The stage being scoped and hired for. Nineteen steps, the same nineteen every lesson. Two of them are done by the software; two are where the software must be overruled.

Prepare
  1. 01Claim the lesson in the app so nobody else picks it up
  2. 02Download the four files from that lesson's Drive folder
  3. 03Group split segments back into correct lesson order
Sync & switch
  1. 04Sync all three angles to the dedicated audio by waveform
  2. 05Mute all three camera audio tracks — the dedicated recorder is the only voice
  3. 06Build the multicam timeline
  4. 07MULTICAM SMARTSWITCH — the software takes a first pass at choosing between three cameras
  5. 08Watch it back and correct every camera choice the software made
Edit
  1. 09Trim the head and the tail
  2. 10TEXT-BASED EDITING — strip false starts, repeated words, excessive “um” and “uh”
  3. 11Never remove silence — see below
Polish
  1. 12Clean the audio: cooler hum, room noise, level balance, voice clarity
  2. 13Match colour and exposure across all three cameras
QC & deliver
  1. 14Watch the full lesson start to finish — not scrubbed, not sped up
  2. 15Sign the fourteen-point QC checklist
  3. 16Picture lock — no more cuts
  4. 17Render the master. Once.
  5. 18Upload three deliverables and write the delivery note
  6. 19Mark the lesson delivered in the app
The rule that matters most

Silence is not dead air. It is the lesson.

The designer regularly works with her hands for thirty or forty seconds without speaking, and that is precisely what a student paid to watch. Every editing tool has a “remove silence / tighten this up” automation, and on this footage it destroys the product while looking like good work.

Remove speech errors. Never remove speech gaps. When unsure, keep it — an extra half-second of quiet has never been rejected here; a missing demonstration step always is.

Never in scope

Titles · overlays · captions · lower thirds · logos · music · graphics · intros · anything in any language.

All of it is added downstream, once per language. Added here, it has to be stripped out and the master re-delivered clean.

Three files delivered to Drive · status master_delivered
STAGE 3

The gate Abel · the only human review

One touchpoint, and it is the one that has to happen anyway before a lesson goes further. There is deliberately no draft-approval round beforehand — the editor's signed checklist replaces it, which is why that checklist is as long as it is.

Accept → the lesson is picture-locked and everything downstream begins.
Reject → timecoded notes go back to the editor. Two rounds are included in the flat rate; the app counts them so nobody has to remember.

↺ On rejection, work resumes at step 04 — not at the render
LOCK
The pivot point

The clean master

FM-0042__bridal-bouquet-mechanics__clean-master.mp4

Never render from a render

One video source · every language derives from this file, forever
Stages 4 & 5 · not built

One master in, one finished video per language out

Each render job takes the overlay-free master, one audio track, and one set of on-screen text — and burns in exactly one language. English is simply the first language, not the parent of the rest.

Why this changed

The earlier plan built each language on top of the finished English video.

That cannot work: the English render already has English text burned into the picture, and no amount of layering puts Spanish cleanly on top of it. Deriving every language from the overlay-free master means each render burns in exactly one set of overlays — its own.

Three things follow. Adding a language later is one new render job — nothing upstream re-runs. Every render is independent, so a failed Spanish job has no effect on French. And dubbing bends audio to fit the locked picture, never the reverse.

Render job 1

English

  • clean master
  • english-clean-audio.wav
  • English overlay text
  • output spec
Render job 2

Spanish

  • clean master
  • AI dubbed audio, timed to picture
  • Spanish overlay text
  • output spec
Render job 3

Portuguese

  • clean master
  • AI dubbed audio, timed to picture
  • Portuguese overlay text
  • output spec
Render job N

Every language after

  • clean master
  • AI dubbed audio, timed to picture
  • translated overlay text
  • output spec
Each job renders on AWS Fargate → S3 destination bucket
The cost of this shape

The master is now a single point of failure for the entire catalogue.

If the master is wrong, every language is wrong. That is exactly why the editor adds nothing to it, and why the one review gate at Stage 3 must not be waved through.

What the software tracks

The status ladder

Where any lesson sits, at any moment. Everything up to ready_for_production exists today; the four after it are what gets built for the editor stage.

creating awaiting_uploads needs_tagging ready_for_rename renaming ready_for_production in_editing master_delivered picture_locked
Built today New — the editor stage Terminal · handed to the render pipeline Rejection sends master_deliveredrevision_requestedin_editing
Second gap found in the code · fixed first

Any new status here would have been silently erased.

The app recalculates a lesson's status every time it re-checks the files in Drive, and it currently protects only one value. An editor could be three hours into a lesson and have it reset to “not started” — their claim gone, the lesson back in the queue, and no error message anywhere.

This is the first thing that gets fixed, before anything else is built.

The handoff

What the editor delivers

Three files per lesson, back into the same Drive folder, named to sort beside their sources.

FileSpecificationWhy
…__clean-master.mp4 H.264, source resolution, high bitrate. Roughly 2–4 GB per hour. Re-encoded on every language render anyway, so mastering-codec quality is thrown away — while a 60–100 GB file would be a multi-hour upload from a remote connection, every lesson.
…__english-clean-audio.wav 48 kHz / 24-bit. Exactly the same length as the master. Used for timing downstream. A one-frame difference is a defect, and it is invisible to the eye.
…__project.drp DaVinci project file. Not a .dra archive. An archive bundles every source file — 100+ GB duplicating footage already sitting in Drive.
The arrangement

How the hire works

Settled

Working terms

Rate
Flat per lesson. Not hourly — render time is inside the price and is never billed separately.
Volume
5–7 lessons per week. A full-time role, not occasional freelance.
Machine
The editor's own. Renders happen there.
Software
DaVinci Resolve Studio, licence bought for them. The free version genuinely cannot do this job.
Revisions
Two rounds included, re-render included. Beyond that, billable.
Review
One, by Abel, on the delivered master.
Access
Per-lesson folders only, through the app.
Ask before hiring

Screening questions

  • “Some lessons have forty seconds with no talking. What do you do with that?” → Leave it. Any answer involving tightening or trimming is disqualifying — they will ruin masters on autopilot.
  • “How long after SmartSwitch runs is the edit finished?” → It isn't. A manual correction pass follows. Treating step 07 as the finish line means they have priced the job wrong.
  • “Which audio do you use?” → The dedicated recorder. Camera audio is for sync and gets muted.
  • “How many times do you render?” → Once, after self-QC.
  • “Studio or free Resolve?” → Tests whether they know the difference.
The one thing left to do

Edit one real lesson personally, following the written procedure exactly.

Not to become an editor — to find out how long a lesson actually takes, which written steps turn out to be ambiguous when a real person performs them, and how good the automatic camera-switching really is on this specific footage.

At five to seven lessons a week, a guess that is an hour out costs a working day every week for as long as the arrangement lasts. That number is the most valuable unknown left in the project, and a fixed price cannot honestly be set without it.