Kobo
Draft   Internal

Video clipping and repurposing system.

Kobo internal use only Draft Client agnostic

01

Opportunity

A video clipping and repurposing system, built once, resold in multiples as a subscription SaaS service.

Long form video in, reviewed short clips out, ranking that improves from approvals and results. Three inputs change per client: the definition of a good clip, the domain word list, and the channels. Everything else is shared.

Four things no tool sells

A client can buy the parts themselves. Several subscriptions, a person carrying files between them, and none of the following.

Their own definition of a good clip

Ranking built on what the client says matters.

Every tool on the market ranks for general engagement. That surfaces what plays to a broad feed, which is the wrong question for any specific audience.

Domain vocabulary

Specialist terms transcribed correctly.

Generic models mangle the words that matter most in a specialist field. The word list is built with the client and fed to the transcription layer.

A review gate

Nothing publishes without a person approving it.

Consumer tools post on a schedule. Any client with expertise or sensitive material at stake needs a person in front of the publish button, and a record of who approved what.

A system that learns

Approvals, rejections, and results change what gets picked next.

Off the shelf tools pick the same way in month twelve as in week one. Nothing they do is informed by which of this client's clips actually worked.

02

Applications

CandidateFitStatus
Doro Mind Strong. Regulated, specialist vocabulary, review requirement. Proposal drafted. Three models. Pricing pending.
SBS Unverified Content repurposing already underway. Scope and overlap unknown. Confirm before assuming anything
The podcast Clean test bed. No regulated data, no compliance owner, no client waiting. Not scoped
Kobo internal Sales and marketing assets from existing recordings Same system, no separate build
03

Tool stack

ElevenLabs Scribe

Buy

Speech to text. Word level timings, speaker separation.

Keyterm prompting takes up to 1000 domain terms, so specialist vocabulary comes back spelled correctly. Reads video directly. $0.22 per audio hour, so effectively free at this volume. HIPAA via signed BAA, plus zero retention and EU residency modes. Verified

Ranking

Build

Ranks moments against the client's own criteria.

The product. Every vendor ranks on general engagement, which is the wrong question for a specialist audience.

Flagging

Build

Marks anything sensitive before review.

What makes the system usable by clients with something at stake.

ffmpeg

Build

Cuts clips from the ranked timecodes.

No clipping vendor accepts supplied timecodes, Vizard included. Cutting from our own is what keeps ranking authoritative rather than a filter on somebody else's shortlist. Verified

Submagic

Buy

Captions, filler removal, per channel finishing.

Takes a video URL plus a custom dictionary. Fed pre-cut clips, so its plan length caps of 2, 5, or 30 minutes never bite. Project creation is rate limited to roughly 30 an hour, which a large backfill will hit. Verified

Review screen

Build

Approve, edit, or reject with a reason.

The rejection reason is what feeds the learning loop. Skipping it is cheap now and expensive later.

Orchestration

Build

Runs the sequence, handles failure, routes files.

n8n works for a client-owned instance. For a hosted product it needs a commercial agreement or a swap. See section 09.

Scheduler and analytics

Theirs

Posts approved clips, reports performance.

Distribution stays the client's. Needs an API so results read back into the loop.

Assessment Every tool assessed 13 tools, verified August 2026
ToolDoesAPITakes our timecodesVerdict
ElevenLabs ScribeTranscriptionYes Verifiedn/aIn
ffmpegCuttingLocal binaryYesIn
SubmagicCaptions, finishingYes VerifiedTakes pre-cut clipsIn
Opus ClipTranscribe, select, captionBusiness tier, custom price VerifiedNoTrial, only if it beats three subscriptions
quso.ai, formerly Vidyo.aiSame, plus schedulingReportedUnconfirmedTrial
VizardSelects and cutsYes, entry tier VerifiedNo. No timecode parameter exists.Out. It would own the selection, which is the product.
Munch StudioDone-for-you serviceNoneNoOut. A competitor. $38 to $398 a month, 5 to 15 brands. Verified
RiversideRecording, plus its own clippingBusiness plan onlyNoOut of the pipeline. Ingest side, upstream of scope.
Repurpose.ioCross-postingYesn/aOut. Distribution is the client's.
FlowjinAudio-first clippingNone verifiedNoOut. Weaker on long files.
ChoppityClippingNone verifiedNoOut. Duration caps.
MontageClipping with its own rankingNone verifiedNoOut. Its ranking duplicates the build.
DescriptEditing suiteLimitedNoOut. Human editing tool, not a pipeline component.

Verified against vendor documentation. Vendors change pricing, gate APIs behind higher tiers, get acquired, or discontinue self-serve products. Munch did the last one. Assume a row here is wrong within six months and re-verify before quoting.

04

System

Seven stages. The client produces and distributes. Kobo builds the middle.

CLIENT PRODUCES AND DISTRIBUTES. KOBO BUILDS THE MIDDLE. 01 INGEST Video in Their recording 02 TRANSCRIBE Text layer Domain terms 03 SCORE Rank moments Client criteria 04 CUT Render clips Reframe, caption 05 REVIEW Approval gate Nothing auto posts 06 PUBLISH Schedule, post Their call 07 MEASURE Performance Per clip BUY BUY BUILD BUILD + BUY BUILD THEIR TOOLS THEIR TOOLS Watched folder Scribe Ranking engine ffmpeg, Submagic Approval queue Their scheduler Their analytics WHAT PERFORMED FEEDS WHAT GETS CUT NEXT THE SYSTEM. BUILT ONCE, CONFIGURED PER CLIENT. DISTRIBUTION. THEIRS. THE SEQUENCE Kobo writes the sequence. Where it runs and who owns it is set by the business model. WHAT THE CLIENT ACTUALLY DOES Shoot as they do today. Drop the recording in. Come back to captioned, reframed clips with suggested copy. Approve the right ones and post on their own schedule. The rest go no further.
05

Wiring

The tools do not talk to each other. Every join is glue that does not exist until someone writes it, and the glue is most of the build.

Why the pipeline is shaped this way

Cutting a clip means knowing where it starts and ends in the recording. Every clipping vendor works those out for itself and none of them will take a list from anywhere else. Ask one to cut a specific moment and there is no way to tell it which moment.

So if the ranking is going to decide what gets cut, the cutting has to happen on our side. ffmpeg does it from a start and end time, which is a small piece of work and the reason the ranking layer is the thing in charge rather than a filter applied to a vendor's shortlist.

It has a second benefit. The captioning service only ever receives clips that are already under a minute, so its per-video length limits never come into play.

What passes between steps

ONE RECORDING IN. APPROVED CLIPS OUT. 01 Trigger A new recording lands in the watched location BUILT 02 Fetch Media pulled down and a job record opened BUILT 03 Transcribe Audio to text, with the domain word list attached BOUGHT 04 Store Transcript saved as a permanent record BUILT 05 Score Moments ranked against the client criteria BUILT 06 Flag Anything sensitive marked for review BUILT 07 Cut Top ranked moments cut to vertical and square BUILT 08 Finish Captions burned in, filler and dead air removed BOUGHT 09 Draft copy Post copy written per channel from what was said BUILT 10 Approval gate Operator approves, edits, or rejects with a reason BUILT 11 Release Approved clips and copy pass to their scheduler THEIRS 12 Measure Performance read back and attached to the job record BUILT 13 Learn Rejections and results adjust the ranking next time BUILT

Two things worth keeping

The transcript. Almost every later step reads from it. The ranking works out which moments matter by reading the text. The start and end times come from it. The draft post copy is written from it. So it gets saved and kept, not deleted once the clips are made.

The job record. One record per recording that follows it all the way through: which moments were picked, what they scored, what got flagged, who approved or rejected what and why, where the finished clips went, and later how each one performed.

Without that record nobody can answer why a clip was cut, who signed off on it, or whether it worked. It is also what the ranking learns from, since improving means comparing past decisions against what happened next.

06

Pricing logic

ComparablePriceWhat they actually get
DIY clipping tools$15 to $100 a month MarketEngagement-ranked clips, no domain vocabulary, no review, no memory
Munch Studio$38 to $398 a month, 1 to 15 brands VerifiedDone-for-you service. Templated output, 500 to 7,500 minutes of repurposing.
Freelance editors$50 to $500 per clip MarketHuman judgment, no system, cost scales linearly with volume
Clipper networks$0.02 to $0.10 CPM MarketIndependent editors posting from their own accounts. Unavailable to anyone regulated.
Managed agencies$1K to $10K a month MarketA team cutting clips for your channels. Forever.
Comparable automation builds$15K to $50K+ MarketComplex orchestration with AI and custom logic
Post-build support plans$200 to $500 a month MarketUpkeep only

Sources: FlowEngine and AY Automate agency guides, Clipping Culture and LuvKaizen service reviews, all 2026. Munch Studio from its own pricing page.

Where the price sits

The floor is $50 tools. The ceiling is $10K a month agencies, paid indefinitely. Model C at $2.5K to $3.5K a month sits between them. The gap above the floor is carried by the four things in section 01.

Cost per clip

The frame that lands hardest against freelance rates. At $2.5K a month Derived:

Clips reviewed per monthEffective cost per clipAgainst $50 to $500 freelance
10$250Mid range
20$125Below most freelance rates
40$63Near the floor
60$42Below it

Volume is the variable. A client feeding the system regularly gets a cost per clip no freelancer can match, and one recording a quarter does not. Worth measuring in Phase 1 rather than promising.

Munch Studio at $398 a month covers 15 brands, which sits well under the agency band and is easy for a prospect to find. What it sells is templated output: no domain vocabulary, no review gate, no learning, and no system the client controls.

07

Business models

Model A   One-time Build and hand off $25K to $40K fixed Market
DORO MIND DORO MIND Shoot the session Unchanged. Patrick's. The pipeline Transcribe, score, cut, caption, queue RUNS ON DORO MIND SERVERS Approve and post Their call, their channels KOBO Builds it, steps back TWO FIXED PAYMENTS, ONE PER PHASE BUILD, WEEKS 1 TO 10 DORO MIND OWNS AND RUNS IT. SUBSCRIPTIONS HELD DIRECTLY. HANDOFF UPKEEP OPTIONAL, QUOTED SEPARATELY

Fixed price per phase, ownership transfers at handoff. Phase 1 alone runs $5K to $8K, upkeep after is optional at $200 to $500 a month Market.

  • Clears every blocker in section 09
  • Client holds the subscriptions, so their footage stays in their own vendor relationships
  • Revenue ends when the build does. The code stops earning.
Model B   Financed Financed build $2K to $3K a month, 12 to 18 months Derived
DORO MIND DORO MIND Shoot the session Unchanged. Patrick's. The pipeline Transcribe, score, cut, caption, queue RUNS ON DORO MIND SERVERS Approve and post Their call, their channels KOBO Builds it, holds admin until paid SAME MONTHLY FEE, 12 TO 18 PAYMENTS, THEN NOTHING KOBO RETAINS ADMIN. RUNS ON THEIR SERVERS THE WHOLE TIME. DORO MIND OWNS IT FINAL PAYMENT, OWNERSHIP TRANSFERS

Little or nothing down, then a fixed monthly fee. Runs on the client's servers from day one. Ownership and admin transfer at the final payment.

  • No-upfront close plus a year of predictable revenue
  • No licensing or custody work, because it never runs on Kobo systems
  • Collection risk. Terms need a clause covering what happens if payments stop.
Model C   Subscription   Target model Subscription on Kobo systems $2.5K to $3.5K a month, no end date Derived
DORO MIND DORO MIND Shoot the session Unchanged. Patrick's. The pipeline Transcribe, score, cut, caption, queue RUNS ON KOBO SYSTEMS Approve and post Their call, their channels KOBO Hosts, operates, supports it MONTHLY FEE, NO END DATE KOBO OWNS, HOSTS, AND OPERATES IT. NO HANDOFF. MEMBER FOOTAGE ENTERS KOBO CUSTODY. REQUIRES DATA TERMS AND AN N8N SWAP OR LICENSE.

Nothing upfront, monthly fee with no end date, clients reach the system through Kobo. This is the only model where several clients share one build.

  • Highest monthly rate, and the only one still earning past the build
  • The engineering becomes an asset a second client can buy
  • Kobo takes on uptime, support, vendor breakage, hosting, and churn
  • Every blocker in section 09 has to be solved first
08

Money over time

MONTH 0 6 12 18 24 A. Build and hand off RUNS ON THEIR SERVERS TWO FIXED PAYMENTS, ONE PER PHASE UPKEEP, OPTIONAL, QUOTED SEPARATELY BUILD DORO MIND OWNS IT HANDOFF KOBO OBLIGATION NONE, UNLESS UPKEEP IS RETAINED B. Financed build RUNS ON THEIR SERVERS SAME MONTHLY FEE, 12 TO 18 PAYMENTS KOBO RETAINS ADMIN DORO MIND OWNS IT FINAL PAYMENT, OWNERSHIP TRANSFERS KOBO OBLIGATION ENDS AT TRANSFER C. Subscription RUNS ON KOBO SYSTEMS MONTHLY FEE, NO END DATE KOBO OWNS, HOSTS, AND OPERATES IT KOBO OBLIGATION UPTIME, SUPPORT, VENDOR BREAKAGE, HOSTING, AND CHURN. NO END DATE. PAYMENT TO KOBO OPTIONAL DORO MIND OWNERSHIP KOBO OBLIGATION
ModelCollected by month 24After month 24Kobo obligation
A. Build and hand off$25K to $52KUpkeep only, if retainedEnds at handoff
B. Financed build$24K to $54KNothing. They own it.Ends at transfer
C. Subscription$60K to $84KContinues, minus hosting, license, and support costsNo end date
09

Blockers

What stands in the way of each model. Everything in C's column is solvable, and all of it is cheaper before the first build than after.

IssueABC
Orchestration licensing. n8n is free for internal use; hosting it and charging for access needs a commercial agreement. Clear Clear Blocked until n8n is licensed or swapped for Activepieces, Temporal, Kestra, Airflow, or owned code
Client footage custody. Anything regulated raises this immediately. Clear, their servers Clear, their servers Open, data terms per client and a BAA where health data is involved
Multi-tenancy. Client separation on shared infrastructure. Not needed Not needed Required, and only cheap if designed in before the first build
The ownership pitch. The client ends up owning the system. Holds Holds, one year later Gone, replaced by low risk and cancel-anytime
Firm model. Kobo sells fixed-scope engagements, not open-ended retainers. Clear Clear, fixed term Departs from it. A deliberate change of business model.
Operations. Uptime, support hours, vendor API breakage, churn. Theirs Theirs at transfer Kobo's, indefinitely, across every client at once
Exit path. Export and deletion, documented. Not needed Not needed Required, or the lock-in objection lands
10

Sequencing

Doro Mind is the live deal, so it sets the pace. Two other tracks can run alongside it.

TrackWhat happensWhy now
Doro Mind First paying build. Model decided before signature. The only track with a buyer and a date
SBS Evaluate for fit and overlap with the repurposing work already underway. Possible parallel build. A second client is what turns a build into a product, and the assessment costs nothing
The podcast Test bed on own material. Same system, no client waiting. Failures land somewhere harmless, and it becomes a worked example later

The model decision

PathTrade
Doro Mind as client one of a productSection 09's work happens now, on a live deal, with the ownership pitch rewritten
Ship Doro Mind under B, productize afterLower risk, keeps the code, delays the recurring revenue. Same asset in the end.

SBS changes this calculation. If it evaluates well and lands close behind Doro Mind, the product case is made by two clients rather than argued from one.