169 turnaround sheets, cut into Instagram carousels by the site's own pipeline
Thirteen years of sculpting left me with hundreds of turnaround sheets — multi-view renders of every figure and ship. Publishing them on Instagram means cropping each view into a carousel slide by hand, which is exactly the kind of chore that guarantees it never happens. The intake pipeline for this website already knows how to find the subject inside a sheet. Could the same analysis cut carousels automatically?
The expensive part was already paid for. Background sampling, panel detection, subject bounding boxes — the site build needed all of that months ago, and the carousel tool is a thin layer on top. The genuinely new problem was editorial, not technical: deciding per view whether the slide should *contain* the whole figure or *cover* the frame with a close-up crop — and a small heuristic (the last view of a figure sheet, when it's clearly wider than its siblings, is a close-up) gets it right often enough that I only override the exceptions.
The tool is done; the bottleneck moved to where it belongs — me. 169 sheets are cut and waiting in an inventory spreadsheet where I mark what gets published and how it's credited. Curation is a human job, and this time that's by design.
The chore, quantified
A turnaround sheet is one wide image with four to six views of a figure. Instagram wants 1080×1350 portrait slides, one view each, consistent framing across the carousel. Done by hand that's five crops, five reframes and an export per piece — multiplied by 169 sheets across characters, units and ships. Nobody does that chore. The work stays in folders.
Standing on the intake pipeline
When this site was built, every sheet already passed through an intake script that samples the backdrop color, detects the panels, and finds the subject's bounding box inside each one. The carousel tool reuses that analysis wholesale and adds only the slide logic:
- each view becomes one slide, in sheet order, at a shared aspect ratio;
- a numbered contact sheet per piece — all slides in a row — so I can review a carousel in one glance instead of opening nine files;
- edge cases the sheets themselves taught us: two views touching get split at the thinnest column of background pixels, slivers get merged back into their neighbor.
Contain or cover: the editorial heuristic
The one decision per slide that matters visually: does the whole subject fit with padding (contain — right for full figures), or does the slide fill and crop (cover — right for close-ups)? Marking that by hand across 800-odd views would be the same chore again wearing a different hat.
The pattern in my own sheets provided the rule: figure sheets are columns of tall views, and when the last view is clearly wider than the others, it's a detail close-up — give it cover. A sheet of wide views throughout is a ship or a diorama — contain everything. That single rule reads the sheet the way I would, and the flags exist for the exceptions.
Where the human went
The tool didn't remove me from the process; it moved me to the only step that needs me. Alongside the exports lives an inventory spreadsheet — every piece, one row — where I mark what gets published and how authorship is credited (my sculpts for Firelock sit next to my own IP, and those need different captions). The machine cut 169 sheets in minutes. Deciding what the world sees, and in what order, is still my job — and the whole point of automating the rest was to finally have the time to do it.