Adam Shankman reverses his AI claim: “ZERO shots” becomes “some windows”

Director Adam Shankman has now flipped the script on his own AI denial for Stop! That! Train!, conceding that some of the film’s train windows were AI-generated. In an interview with Xtra Magazine conducted before he made the public denial, Shankman says “some of the windows” and “one or two” shots of the train were generated using AI.

This matters because Shankman previously framed his position as a clean break: in a statement after the denial, he said “there are a sum total of ZERO shots conceived by AI in the movie.” The A.V. Club also notes that critics called that phrasing “intentionally vague,” and the new, more granular admissions appear to directly contradict the earlier absolute.

According to Shankman’s Xtra interview, the specifics are surprisingly concrete. He tells Xtra: “There was only one AI shot or two AI shots of the train, the rest are CG.” Then comes the key detail for anyone assessing how “real” the AI use was: “Most of the CG is the train, then out the windows is mostly AI.” Shankman also claims that the Florida station is entirely CG, adding: “And different licensed imagery.”

Bleecker Street, the film’s distributor, expanded the picture in a statement to Xtra. Their version is that AI tools were used by the VFX team to assist in three moments: (1) to adjust the angle of live action footage outside the windows to match the camera angle in one shot; and (2) to format the CGI-crafted train onto 2 frames of live action footage exteriors. Bleecker Street also emphasizes that traditional CGI was used, including for the train and the stormaganza, and that “hundreds of VFX artists were hired and paid” for the film. In other words, the dispute is not “was any work automated at all,” but rather “what counts as an AI shot,” and how honestly that was communicated to the public.

For decision-makers, this is where the story stops being just a film-industry drama and turns into governance homework. Shankman’s earlier statement is described as trying to deflect concern about AI involvement. But the new disclosures, plus follow-up reporting inside the piece, suggest there were visual inconsistencies in trailer VFX shots tied to the train windows. The A.V. Club cites that last week, post-production professional Gloria Cook pointed to the number of windows on the train as an example of AI usage, and to a mismatch between the number of VFX compositors credited (26) and how many CG artists were credited (zero). The inference in the article is blunt: critics’ reading that Shankman and Bleecker Street’s comments were evasive appears consistent with the observed inconsistencies.

The incentives here are obvious to anyone who’s ever managed a launch. AI use is now both a cost lever and a brand-risk mine. In the source, Shankman frames the production as cost-stretched. He says the film was shot in 19 days and describes the creative aim as “Orient Express-meets-Wes Anderson kind of vibe. For 10 dollars,” adding: “We were working with string cheese, not brie.” If you believe his framing, the pressure to cut time and cost is real. But the second-order problem is also real: if cost or speed motivates AI usage, then transparency becomes the reputational “budget line” that can’t be shaved. When a public denial turns into a later admission of “some windows,” trust is what gets eaten first.

There is also a labor-and-community subtext that execs and boards should not ignore, because it shows up in audience sentiment and in who gets credited. Cook’s critique is summarized in the A.V. Club piece as a jobs question: genAI replaced a job that could have been filled by a queer artist, and a queer film aiming to attract queer audiences would plausibly benefit from that kind of creative representation. The article ties this to Shankman’s stated hopes about the moment, while arguing that his attempt to skirt the issue undermines the credibility of his broader claims.

Strategically, the Stop! That! Train! episode is a preview of the next governance pattern. Public denials, vague phrasing, and marketing-language tightropes are no longer just “PR polish.” They become discoverable records that can contradict later disclosures, and they can trigger scrutiny from critics, professional communities, and potentially regulators depending on how AI use is governed in specific jurisdictions. Even if the final legal exposure is not spelled out in the source, the reputational and operational risk is immediately legible: distributors inherit the messaging risk, VFX teams inherit the technical accountability, and directors inherit the credibility hit.

Bottom line: Shankman’s new admission, that “some of the windows” and “one or two” shots were generated using AI, turns a supposed “ZERO shots conceived by AI” narrative into a more complicated reality. For executives across entertainment, that is the real takeaway. Your AI controls can be technical and your workflows can be defensible, but your public story needs to be as precise as your post-production pipeline. Otherwise, the next trailer inconsistency becomes the next corporate crisis.