Amplitude now has to prove it filmed its own trailer

Gizmodo's specific complaint is a five-second stretch, 0:35 to 0:40, in which a woman sprints toward a camera that backs away from her and stays perfectly or almost perfectly in focus, with motion blur showing up only at the edges of the frame ^[8]. That is hard to buy on a real set: deep focus with real lenses wants a lot of light, and gets harder the moment the camera starts moving ^[9]. Five seconds out of ninety is 5.6 percent of the runtime ^[14], and it appears to be doing the work of judging the other 94 percent.

Viewers actually watch a trailer once, at whatever bitrate the storefront hands them, and file it as filmed or synthetic within the first few seconds, which is why Gizmodo also flags what Steam and YouTube compression does to an image before anyone sees it ^[10]. Marketing teams tell themselves viewers do something different: read the description, then the follow-up post, then the behind-the-scenes gallery, and revise that first impression. That revision rarely travels as far as the original accusation.

Amplitude's answer so far is a materials list and an address, real actors and practical sets at Nu Boyana in Sofia ^[4][5]. Per Gizmodo, the studio has stopped short of saying that no generative AI was involved anywhere in the trailer ^[11], and that gap is where the argument keeps living. A flat denial would face the same limit, since a viewer has no way to check one.

The grid worth drawing has two axes: whether you used generative video, and whether the finished cut reads as AI to a stranger on a phone. Three of the four cells are manageable. Used it and it reads that way: say so in the copy. Did not use it and it reads photographed: nothing to do. Used it and it reads photographed: you have a disclosure problem rather than a perception one. The fourth cell, no generative AI and it still reads synthetic, is the one Amplitude is sitting in ^[12], and it is the only cell where the fix is evidence rather than craft.

The test costs a morning. Cut a compressed version, give it to five people who were not on the project, and ask them what made it. If two say AI, there are two paths and both cost something. Change the image so it reads as photographed, which means grain and a focus puller who misses, and accepting a worse picture by every objective measure ^[15]. Or publish the proof in the same hour as the trailer instead of promising it soon ^[6], which means a stills photographer on the shoot day and clearances arranged before the premiere rather than after the pile-on.

Whoever signs off the cut owns that choice, and the cheap version of owning it is the stills budget. The damage itself is harder to price from outside the studio: the Gizmodo account records the chorus, not wishlist adds or watch-through, and Amplitude holds those numbers alone ^[17].