Garage / Carlsberg · evidence pack · 15 September 2026
Seven keyvisuals and a sticker set, rebuilt from every generated image, every saved working file and every message in the Missouri Creative threads. Prepared as a record of what the work involved. An in-spec round costs hours. A round that changes the logic costs days.
The hour figures are built from the work record, not from a timesheet. Hands-on hours come only from moments production demonstrably acted (a change to a generation setup, a saved 3D scene, texture or render, a saved working or delivery file, a client email sent from production), grouped into working sessions. Where the producer wrote a substantive client reply outside those sessions, an agreed 15 minutes is added for the consultation behind it; the producer's own time is on the Producer tab. Machine hours are the time generation and 3D rendering were running. The two overlap, because machines run while people work, so they are never added together. Retouching between saves does not register, so hand craft is still under-counted.
Calendar bookings are shown for the sticker phase as a dashed reference only. The habit of blocking time started partway through this project, so the calendar is a partial record and is not used to validate anything.
The seven-week gap is summer holiday, not idle project time, and is excluded from every lead-time figure here.
15 April to 25 June. Solid bars are hands-on hours and pale bars machine time; the green and navy ticks below are client mails in and our mails out. Feedback on this phase ran largely through Slack rather than mail, so the mail lane understates the traffic.
Hours and files are counted from the work record. A comment round is a day on which the client asked for changes or made a choice on that image, in email or Slack; calls are not included. Comment load is our own rating of each round, a judgement rather than a measurement.
Most-revised image of the set. Bottle choice, cabinet, crisp packets, a shirt fix, a face upscale and a re-crop of the group each landed as a separate version.
19 May – 16 Jun · incl. 3.2 h 3D
Two design starts: seven skateboard studies in April, then the Cherry build from v1.0. Third-highest hands-on time of the set, on 602 generations.
23 Apr – 3 Jun · incl. 2.2 h 3D
Master document created 15 April, the day the brief arrived. Twenty-seven layers, and an ethnicity-swap test the client asked to see.
28 Apr – 3 Jun · incl. 1.5 h 3D
Steady run. Five comment rounds, then three expansion passes for the wider crop.
11 May – 9 Jun
One named client feedback file on 8 June, then two expansion versions. The headphone base was reworked following feedback on 22 May.
7 May – 15 Jun · incl. 6.1 h 3D
Tightest run of the set: 13.2 hours hands-on despite 2,686 generations. Ten versions, four comment rounds, and the lowest hands-on time of the seven.
18 May – 23 Jun · incl. 2.3 h 3D
Document created 28 May, when the range image entered the schedule. Ran to v2.2 with separate sticker and no-sticker cuts.
2 Jun – 23 Jun · incl. 3.4 h 3DThe Shower Cap took ten versions and 13.2 hands-on hours. The Fridge took twenty versions and 21.3, spread over 19 active days. Same brief, same team, same tools. The difference is how much of the spec was settled at the start.What a bounded brief is worth
10 August to 14 September. The dashed line is calendar-booked Garage time where it exists; the grey footing is other projects booked that day. Note the work either side of the 9 September approval.
Six delivered batches. The first was one sticker per keyvisual: the existing artwork adapted, labels corrected and upscaled. Everything after that was new design.
One sticker per keyvisual: existing artwork adapted, labels corrected, upscaled. Created 11–14 August, out on the 12th. Chair, king, boa, DJ, bath and skateboard. Cheap per sticker precisely because the artwork already existed.
The family that drove round five. Pattern logic was rewritten three times: keyvisual-derived, then per-SKU colour, then pattern-per-ABV-pillar.
Sent unspecced as a sample on 2 September. The spec was settled in the response that followed: background per range, motifs from label artwork only, corner graphics dropped.
Round two was entirely typographic: angle, kerning, %ABV position, saturation. Resolved when Jonny supplied his own lockup on 1 September, placed back inside the flames, colours matched to the approved 6% and 8%. Shown here as delivered, not as the bare lockup.
Four products, one marbled treatment each, built 3 September once colour-per-SKU was settled. 419 generations that day alone.
Built 9 September on the settled structure. Briefed Friday, delivered Wednesday, approved 42 minutes later.
Same window as the halftone set. Flame colour keyed to the Hardcore SKU, a lookup rather than an exploration.
Round five was the largest round on both measures: 16.3 hours hands-on and 17.3 hours of machine time, more generation than rounds two, three and four together. It is the only round in which the pattern logic itself changed rather than the output.
Every generated image assigned to the keyvisual or sticker it belongs to, and every hour sorted by the kind of work that proves it. The split matters for pricing: machine hours fall as the pipeline improves, hands-on hours do not.
Machine time ran 48.5 hours in the sticker phase, but the client's comments were almost never about generation. They were about kerning, hue, a keyline that had gone off-white. That is hands-on work, and it is the part no pipeline improvement removes.Why the two should be priced apart
Every message in the 22 client threads, counted.
Worth knowing which figures survive a challenge and which need a caveat, before any of them are quoted.
Three tiers. Every figure is taken from this project.
A named fix to an approved asset. No new logic, no new assets needed.
17 Aug: Freyja's comments 11:59, new versions out 15:47, approved 16:53. Skateboard reprioritised 18:47, v1.2 delivered 11:46 the next morning.New SKUs or colourways inside a spec that already exists. Volume is cheap here.
Brief to first batch: 36 hours. Fourth batch: 4 working days, 621 files.Changes the pattern logic, the colour source or the layer model. A new project, not a comment round.
Round five: Freyja 25 Aug, then Jonny's 9.9 MB of lockup material on 1 Sep. Spec revision plus a pipeline rebuild.Missouri's total review time across all six batches was 2.2 days. The 4 September batch came back in under three hours; the 9 September one in forty-two minutes.Client review time
Three items still open.
Revised pricing basis, the three turnaround tiers, and wording for the scope ceiling. Kimmo committed to this on 14 September.
A short written note covering the keyvisuals and stickers already made. Oldest open item.
Kimmo has said the commercial frame comes first, so these should follow that answer rather than precede it.
Method: every generated image was assigned to its keyvisual or sticker by comparison with the delivered finals. Hands-on hours come from changes to the generation setup, saved 3D scenes, textures and renders, saved working and delivery files, and client emails sent from production, grouped into sessions that end after 45 minutes without activity, plus an agreed 15 minutes for each substantive producer reply written outside those sessions. Machine hours are the time generation and 3D rendering were running. Client dates from 180 messages across 22 threads, read in full. Images are the latest delivered versions.