Week 12: Slop as a genre, restated

A retrospective on the semester's argument that AI-generated video is a genre with its own tells, economics and ethics, using the case studies that carried it

10 May 2027

Twelve weeks ago this course opened on a claim that sounded like a joke: that AI-generated video deserves to be studied the way film school studies melodrama, as a genre with recognisable conventions, not graded as a technology demo on how close it gets to realism. The semester’s actual argument was never about realism. It was about three things realism has nothing to do with, and it’s worth saying plainly, once, what they were and where the evidence for each one came from.

The tells are conventions, not glitches#

A six-fingered hand, a logo that resolves differently between frame four and frame forty, a voiceover cadence too smooth to have needed a breath — the early weeks treated these as things to notice, and the later weeks explained why they persist: they’re the visible edge of a review process built to catch some failures and not others. Halcyon’s citation error shipped eleven thousand times not because the tell was invisible but because nobody built a gate that looked for it. A genre’s tells are always evidence of what its production process does and doesn’t check.

The economics, not the aesthetics, explain the scale#

Week 9’s arithmetic — cost per render against ad rates per thousand views — is the actual reason slop looks the way it does: cheap enough to produce in volume, and volume is where the margin lives, so review time, the one cost that doesn’t fall with better models, gets spent as thinly as the business can tolerate. That’s not a failure of ambition. It’s the rational output of the cost structure, and it will keep producing the same shape of artefact for as long as the cost structure holds.

The ethics get harder exactly where the craft gets better#

Nightshade & Nectar’s seed drift was a coherence failure anyone could point at. Marlowe Vance was the harder case this course sat with: a synthetic performer good enough that the director’s statement’s job stopped being “explain the tools” and became “account for a likeness nobody explicitly consented to lending.” Better technology doesn’t retire this problem, it raises the stakes on the paratext that’s supposed to disclose it.

None of this is an argument that slop is good filmmaking, and it was never trying to be one. It’s an argument that slop is a genre with a coherent internal logic — tells that mean something, an economics that predicts them, an ethics that gets more urgent as the craft improves — and that a course teaching you to make one, tear one down, and automate one has taught you more about how moving images get made in 2027 than a course teaching you to admire the ones that took a hundred people and a year would have. Whether that’s a good use of a semester is, appropriately, a question this course leaves for your director’s statement, not its own.