Whoever Had to Sign It

An iron screw press clamping a thick stack of blank sheets on a dark workshop bench, while beside it an identical stack has come apart and its blank pages lift and scatter into the air.

Two prices were put on machine-made film and television this month, in two kinds of paper, and only one of them contains any terms.

The first. Utopai Studios raised $60m at a $1bn valuation, and a person close to the company says it is fielding follow-on offers that would value it at $2bn to $3bn, a report in The Information relayed on 25 October. About twenty projects in development, three features scheduled for theatrical release next year, roughly three thousand North American screens — where most AI features so far have opened on one, as demonstrations. The chief executive, Cecilia Shen, formerly of Google X, says that selling software alone is not enough in entertainment. The company says its tools cut scene planning and computer-graphics costs by eighty to ninety per cent against conventional production.

None of those numbers has anywhere to go. A valuation is a sentence about 2027. The cost saving is the company’s own arithmetic on work nobody outside it has seen. The twenty projects are projects.

Exactly one number in the announcement has an outside referent, and it is worth sitting with. Utopai X, the studio’s video model, sits second on Artificial Analysis’s text-to-video leaderboard: Alibaba’s Wan 3.0 at 1174, Utopai X at 1168, MiniMax H3 at 1158. Artificial Analysis reports that Utopai X is post-trained on MiniMax H3 — the model ten points below it. And the board gives Wan 3.0 a rank of “1-2”, declining to separate first place from second. So the single independently measured fact about the self-described world’s largest independent AI-native film and television studio is a placing that the measurer will not quite award, on a model derived from someone else’s, behind one built by a conglomerate.

I do not offer that as an embarrassment. Post-training an open model is a sensible thing for a studio to do. It is simply the only part of the week’s paper a stranger can check.

The week’s second deal is a distribution one. Wonder Studios agreed with Foundation Distribution to put AI-made work on connected television, Roku and Samsung among the platforms named, beginning with the second season of its anthology Beyond the Loop, which is already on YouTube (C21Media, 22 October). Foundation’s president, Sarah Craig, says that bringing creator work to connected TV “doesn’t dilute existing audiences”. Wonder’s chief executive and co-founder, Xavier Collins, calls the agreement “a watershed moment for Wonder”. The coverage gives no revenue split, no term, no territory, nothing about rights and nothing about credits. The announcement called the deal first-of-its-kind. The trade paper that wrote it up did not repeat the phrase.

Now the other kind of paper. SAG-AFTRA’s members are voting on a four-year Television Animation Agreement, and ballots close on Thursday at 5pm Pacific. Three per cent in each of the four years, the first retroactive to 5 July 2026. An average five per cent increase in residuals when a programme moves to subscription streaming. The subscriber factor for qualifying programmes on the largest foreign streaming services goes from thirty-nine per cent to forty-one. Health plan contributions up a per cent, pension up a per cent from 2028, and a further quarter of a per cent to fund a new paid parental leave programme (Cartoon Brew, 6 October).

The agreement being replaced is still up on the union’s own site, and it is the document this week’s money has no equivalent of. “Voice actors” includes “only humans”. Consent is required when a performer’s name is used in a prompt. A producer creating any synthetic voice through generative AI must give notice and bargain over appropriate consideration. A performer whose work is digitally altered into another language becomes eligible for residuals.

A per cent here, a quarter of a per cent there, against a billion dollars and three thousand screens. The asymmetry is the obvious thing about this and not the interesting one. What holds me is which of these two kinds of paper a historian will be able to use.

We know what fifteenth-century workshops made, and roughly what it paid, because somebody had to sign for it: a commission naming the pigments, a date for delivery, a penalty for missing it. We do not know what those workshops believed about themselves, because nobody wrote that down in a form anyone was obliged to keep. The surviving evidence of any period of making is the paper that bound somebody. A valuation binds nobody. A watershed moment binds nobody. Forty-one per cent will still say forty-one per cent in 2040, and anyone will be able to look up who owed it to whom.

Which makes the last figure the one I would watch. When SAG-AFTRA ratified its TV/Theatrical agreement this year the result was 91.42 per cent in favour — on a 19.25 per cent return. The only paper in this story with enforceable terms in it is also the paper that four-fifths of the people it covers did not vote on. Both prices were set this month. Only one of them had to be agreed to, and most of the people it will bind are leaving the agreeing to somebody else.


Sources