MRCL MAKE — Magnum adaptation & localisation studio
01
Brief intake
Parse the brief → production plan · variant matrix · compliance
Connected ID
MRCL MAKE · Human x AI craft studio
One baked master in. Every flavour, every market, out.
Human x AI craft — deconstruct, re-skin, localise.
Parsed brief — TMICC AI Test
ClientMagnum Ice Cream · standalone (ex-Unilever)
Brief IAI Product Demo Sequence Variant Adapt
AdaptLa Pistache → La Pêche · 3 x key shots, 2″–3″ each (bonus: another flavour/product variant)
Brief IIVO Versioning — Global English Master → local VO in FR · DE · TR · ES, neutral accent
SourceHD Masters · Brand World · TV Campaign Masters
ComplianceEU AI Act · 🇫🇷 real-food rule flagged
3 shots4 VO langs · neutral accent1 flavour swapEU residency
What changes for the team
CostMonths of render time collapse to days
ProcessOne flow, brief to delivery — no handoffs
SpeedRe-skin the layers, don't re-run the sim
GovernanceOne source of truth your team can sit inside
Tech innov.A connected core, not another isolated tool
DataCost and time tracked at every step
Cost & time matrix — proposed, illustrative
CG demo (today)~3 months · Houdini/Nuke
MRCL MAKE~6 days · −54% · est. €2m+/yr saved
Gen AI generation of product variant demo sequence/film from an existing masterProposed: per-shot rate, illustrative
Gen AI-generated VO adapt from an approved AI VO masterProposed: per-language rate, illustrative
How this actually starts — upload → deconstruct → adapt
1 · Upload brief in platform
Brief + reference assets land in MRCL MAKE
● Platform intake
References uploaded: HD Masters · Brand World · TV Campaign Masters.
Step 1~1h
The step that makes everything else possible
2 · Deconstruct
Baked flat master → editable layers
◆ Manual team — honest, not automated
The master lands as one flat render with nothing to pull apart. Our team takes it apart by hand — background, product geometry, coating sim, copy, logo — into the same recovered layers you'll see in Workspace 02. Do this once and every future variant is a re-skin, not a re-build.
Step 2~set-up, then reusable
The engine is now loaded with the supplied July 1 sample corpus: Flavour Remix Examples, CGI / AI Driving Examples, and Backgrounds Without Product. The clips below are clean local copies with safe filenames, so the workflow shows the real material MRCL MAKE can deconstruct, remix and route.
Ingested assets — real, supplied
La Pêche — flavour remix outputs Real, supplied
From _FlavourRemixExamples · the three key MRCL MAKE test shots: crack, dip and pour.
01 Crack
02 Dip
03 Pour
CGI / AI driving examples Real, supplied
From _CGI_AI_DRIVING_EXAMPLES · product drive, no-product pass and background fall plates.
01 Product drive A
02 Product drive B
03 No-product plate
English master mix + splits Real, supplied
Page 5's "Master VO Adapt Example" + "Sound Mix and Splits" — the source we localise from, not the finished FR/DE/TR/ES output.
01 Full mix
02 VO split
03 Music split
04 FX split
Also ingested: product-free background plates, including bg-without-product-runway-kling-6.mp4 and the full HD Runway/Kling plate for re-composition.
02
Deconstruct
Recover the baked master into editable layers
5 layers
Recovered layers — toggle to isolate
Background / environment
Matte + depth relight
AI auto
👁
Product shape + stick
Geometry roto · clean plate
AI auto
👁
Coating / dip / crack — the hard one
Sim layer · hero motion
→ specialist
👁
Copy / motion graphics
Flavour name · claims
AI auto
👁
Logo
MAGNUM wordmark
AI auto
👁
5layers recovered
4AI-automatable
1route to specialist
The honest part: the coating/crack sim is the hard 1%, and we don't fake it. It routes to a specialist simulation or manual Houdini team (Workspace 5) with an auto-generated brief attached — while the other four layers re-skin instantly.
🇫🇷 France hero shot→ Manual team · live-food shoot
03
Variant studio
Swap flavour — re-skin every layer off one geometry
4 flavours
Target flavour
La Pistache
Master
La Pêche
Test brief target
La Framboise
Bonus variant
Caramel
Bonus variant
What actually moves, layer by layer
CoatingPistachio green → Peach
BackgroundOrchard palette shift
Copy“La Pêche” · FR-set
LogoMAGNUM · unchanged lockup
GeometryReused — zero re-sim
Live variant preview
La Pêche
MAGNUM
La Pêche
The actual output — La Pêche, 30″ Real, supplied
Crack
Dip
Pour
These are the supplied La Pêche flavour remix shots — the true rendered result, not the animated mock-up above it.
Flavour remix — real AI · ComfyUI fluid-sim, control-net driven Real, produced
La Pistache · product drive
La Pêche · base crack
La Pêche · base dip
Background plate · HD
Background plate · 16:9
Fall background drive
Real AI flavour remixes off one geometry — the coating and fluid re-simulated per flavour via a control-net-guided ComfyUI pass. Same bar, new taste, no re-shoot, no re-sim from scratch.
04
Sound & localisation
EN master → FR · DE · TR · ES · music per market
5 markets
VO versioning — ElevenLabs
🇬🇧
English (UK) — Global master Real, supplied
Master
Approved
🇫🇷
Français — neutral accent
+ 🇫🇷 food rule
Done
🇩🇪
Deutsch — neutral accent
neutral
Rendering
🇹🇷
Türkçe — neutral accent
neutral
Queued
🇪🇸
Español — neutral accent
neutral
Queued
The English row is the real supplied master VO split, playable now. FR / DE / TR / ES are wired up and waiting — Matt's localised AI VO assets haven't landed yet, so those players are staged for real renders rather than filled with placeholders. Per the brief, results land sharper once a full phonetic VO-artist pass is layered in — expected at this stage, not a shortfall.
Music — Lyria 3
BedWarm, premium, orchard-summer
Markets1 bed · tempo/instrument variants
MixVO ducked · ffmpeg sidechain
Transcreation — Gemini 3 Pro (literal copy translation for a neutral accent — a creative-translation pass would sharpen this further)
EN“Made of pleasure.”
FR« Fait de plaisir. »
DE„Aus Genuss gemacht.“
TR / ES“Hazdan yapıldı.” · “Hecho de placer.”
Sound mix & splitsReal, supplied
The brief's page 5 reference — "Master VO Adapt Example" + "Sound Mix and Splits." The English master mix and its VO/music/FX stems are the real source we localise from. Per-language splits for FR/DE/TR/ES follow once Matt's localised AI VO assets land — gated on source audio, not forgotten.
01 Full mix
02 VO split
03 Music split
04 FX split
Per-language splitsPlanned — pending localised VO assets
05
Orchestration
Route each node · internal AI · specialist · manual
8 nodes
Master · source
La Pêche
Deconstructed · 5 layers recovered
Connected ID
Layer · internal
Background
Matte + depth relight
● Internal AI~2h · €
Layer · internal
Product + stick
Geometry roto · clean plate
● Internal AI~3h · €
The hard 1% · route out
Coating / crack sim
Hero motion · auto-brief SOW #4471
▲ Specialist sim~2 days · €€€
Compliance-gated · 🇫🇷
France hero shot
Real-food rule · live-food shoot
◆ Manual team~4 days · €€€
Layer · internal
Copy / motion graphics
Flavour name · claims
● Internal AI~1h · €
Localise · route out
VO master + langs
EN → FR / DE / TR / ES · neutral
◇ ElevenLabs~3h · €€
Localise · internal
Music
Lyria bed + per-market variants
● Internal AI~1h · €
Converge
Compose + mux
WEV/Remotion + ffmpeg
● Internal AI~1h · €
Deliver · results
6 days
est. end-to-end · −54% efficiency
7nodes
2briefed out
1manual · FR
06
Review & feedback
Frame.io-style review · comment on the frame · approve
v3EU-hosted
Comments 2 open
Click anywhere on the frame to drop a note at that spot + timecode. Click a note to seek there.
