Skip to content
Self-initiated concept work · 4 builds

Work built where craft meets compute.

Everything here is concept work. Gulp, Keel, Hare and Cendre are brands we invented so we could set our own briefs and stress-test the pipeline end to end — no client commissioned any of it. The numbers on each case describe the build: formats delivered, set days used, turnaround achieved. Bring us a real catalogue and we will show you the same system on your product.

01 / 04Hybrid Brand Film2026

Cendre

Luxury fragrance · DTC

Nine days, no set, full cinema.

9 days
Concept to final cut
1 → 14
Shoot days to hero cuts
3
Worlds built in post

The brief we set ourselves

The constraint we set ourselves: launch a debut scent with a cinematic film on one shoot day, no product photography, and a fixed nine-day window.

The Bark move

We directed a single live-action day for texture and talent, then extended every frame in neural post — building the bottle, the light, and three campaign worlds that never existed on set.

60s brand filmCTV masterPaid social rolloutNeural finishing
Request the build breakdown →
Frames from the buildThree campaign worlds and a bottle that never existed on set.
  • Cendre — World 01 · salt flat

    World 01 · salt flat

  • Cendre — World 02 · marble corridor

    World 02 · marble corridor

  • Cendre — Atomiser, built in post

    Atomiser, built in post

  • Cendre — Product ledge

    Product ledge

02 / 04Performance Campaign2026

Hare

Performance footwear

An ad engine that never fatigues.

120/wk
Variant capacity
11
Worlds from one product
1
Creative spine

The brief we set ourselves

The constraint we set ourselves: forty-plus live creatives fatiguing faster than a team can replace them. Build a system that produces ads, not another batch of one-offs.

The Bark move

We built a modular motion engine — one creative spine, infinite variants. New angles, hooks, and cutdowns ship from templates in hours, so the feed never fatigues and the brand never drifts.

Modular ad systemMotion templatesPaid socialDOOH adaptation
Request the build breakdown →
Frames from the buildOne locked product, eleven worlds — four of them here.
  • Hare — Track, night

    Track, night

  • Hare — Taxi rank

    Taxi rank

  • Hare — Trail

    Trail

  • Hare — Locker room

    Locker room

03 / 04Neural Production2026

Keel

Fintech app

Four markets, one build, zero set days.

0
Physical set days
4
Markets from one build
100%
Synthetic environments

The brief we set ourselves

The constraint we set ourselves: launch imagery for four markets at once, for a category that lives on a phone screen — with no set build and no crew on any of the four coastlines.

The Bark move

Zero set days. We built one quayside, then relit and re-skied it into Cape Town, Rotterdam, Lisbon and Reykjavík — same light, same props, four markets. Interface work was designed for real and composited into every screen.

Neural keyframesMarket localisationRelight & finishUI compositing
Request the build breakdown →
Frames from the buildOne quayside, relit and re-skied into four markets.
  • Keel — Cape Town

    Cape Town

  • Keel — Rotterdam

    Rotterdam

  • Keel — Lisbon

    Lisbon

  • Keel — Reykjavík

    Reykjavík

04 / 04Commerce Creative2026

Gulp

FMCG · beverage

A whole catalogue from one content system.

30+
SKUs from one system
0
Re-shoots per refresh
Hours
New SKU turnaround

The brief we set ourselves

The constraint we set ourselves: a catalogue of thirty-plus SKUs and still growing, on a weekly refresh cadence, at a budget legacy agencies quote into next quarter. The maths doesn't work the old way.

The Bark move

We turned the catalogue into a content system: one template architecture renders PDP video, retail-screen loops, and always-on social for every SKU — refreshed weekly without re-shooting a thing.

PDP product videoRetail screen loopsAlways-on socialWeekly refresh
Request the build breakdown →
Frames from the buildEvery SKU rendered off one template — no re-shoots.
  • Gulp — Coral

    Coral

  • Gulp — Lime

    Lime

  • Gulp — Deep blue

    Deep blue

  • Gulp — Amber

    Amber

  • Gulp — Magenta

    Magenta

  • Gulp — Teal

    Teal

The Lab · 2 tests

Technique tests, not campaigns.

Between builds we run tests that answer one question each. No brand, no brief — just the hardest part of a job isolated until it either holds up or it doesn't. These are the ones that held.

LAB 01Relighting · interiors

Day to Night

Can a room shot at midday be relit to evening convincingly enough to survive being held against the original?

How we ran it

Five real interior photographs, cropped once and relit — not filtered. Every pool of light in the evening frame comes from a fixture already sitting in the daylight photograph, switched off. The daylight frames are untouched originals, which is what makes the test worth running: when the before is real, every cheat in the after shows up instantly.

The tell we watch for

A bad day-to-night keeps the window at its daylight value and tints the rest, so you get a white portal in a dark house. The second tell is a night frame with no blacks in it — that is a day frame wearing a filter.

  • Day to Night — 01 · Slat wall — day
    Day to Night — 01 · Slat wall — night
    DayNight

    01 · Slat wall

    The wall lights itself and the hillside outside drops to silhouette.

    Daylight frame: aranprime

  • Day to Night — 02 · Pleated lamp — day
    Day to Night — 02 · Pleated lamp — night
    DayNight

    02 · Pleated lamp

    The lamp was always the key light. Nobody had turned it on.

    Daylight frame: Bilal Mansuri

  • Day to Night — 03 · Yellow armchair — day
    Day to Night — 03 · Yellow armchair — night
    DayNight

    03 · Yellow armchair

    One brass lamp, one direction of shadow, and the chair keeps its colour.

    Daylight frame: Kam Idris

  • Day to Night — 04 · Table and chair — day
    Day to Night — 04 · Table and chair — night
    DayNight

    04 · Table and chair

    A pendant just out of frame, and one mug's worth of light.

    Daylight frame: Ryan Riggins

  • Day to Night — 06 · Green wall — day
    Day to Night — 06 · Green wall — night
    DayNight

    06 · Green wall

    Two sconces, two ovals on a green wall, and the mirror catches both.

    Daylight frame: Spacejoy

Drag any frame to wipe between the two. The daylight halves are real photographs, used under the Unsplash licence and credited to the photographers who shot them.

LAB 02Identity · beauty

One Face

Does a generated face stay the same person when the lens changes?

How we ran it

Two faces, neither of which exists. Each was built as a raking-light plate and then pushed to a macro frame close enough to read pores, lashes and moles. Nothing is retouched — there was nothing to retouch. Anyone can generate a beautiful face; holding one across setups is the part that decides whether you can build a season of work on it.

The tell we watch for

Two catchlights in one eye is how you spot a fake. So is skin that gets cleaner as the lens gets closer — real skin gets more honest.

  • One Face — Face A · plate

    Face A · plate

    Raking side light, no makeup, no fill. Pores show up only when the light is unkind.

  • One Face — Face A · macro

    Face A · macro

    Same woman, same moles, one catchlight — at a focal length that hides nothing.

  • One Face — Face B · plate

    Face B · plate

    A second identity built the same way, to check the first was not a fluke.

  • One Face — Face B · macro

    Face B · macro

    Freckles and vellus hair carried across. The eye is where the pipeline fails first.

On the bench:DTC apparel hero filmFintech CTV systemHospitality launch suitePerformance social engine v2Always-on brand contentOn the bench:DTC apparel hero filmFintech CTV systemHospitality launch suitePerformance social engine v2Always-on brand content

Your turn

Want your next launch built with the Bark edge?

Start a brief →