I’ve watched enough AI video now to know the usual test: can it produce a human hand without accidentally inventing a sixth finger?

X user @socratech has chosen a more rigorous benchmark.

Can they put Jack Sparrow, Neo, the Terminator, the Bride from Kill Bill, the Joker, Indiana Jones, Marty McFly, Mia Wallace, John Wick, Forrest Gump and Walter White into the same suburban hallway — and have them punch one another in the face?

The answer, magnificently, is yes.

The scene is filmed through a fisheye doorbell camera. Jack Sparrow approaches first, presumably to ask why the rum has gone. Neo appears behind him and knocks him flat. The Terminator then terminates Neo. The Bride dispatches the Terminator. The Joker floors the Bride. Indiana Jones deals with the Joker.

You get the idea.

By the time Marty McFly arrives, the hallway looks like Comic-Con has suffered a carbon-monoxide leak.

What makes the video work is not simply that the characters are recognisable. AI has been producing slightly waxy celebrities for a while. It is the continuity of the joke: the same camera, the same room, the same perfectly timed entrance and an increasingly inconvenient heap of Hollywood intellectual property accumulating beside the front door.

MiniMax launched H3 last month as an open-source multimodal video model. It can work from text, images, video and audio, producing clips of up to 15 seconds at up to 2K resolution with native stereo sound. This 52-second sequence has therefore almost certainly been assembled from several generations.

You can see some of the joins. Bodies merge into the furniture. Faces occasionally acquire the smooth, haunted finish of a Madame Tussaud's icon left beside a radiator. The pile of unconscious film characters obeys a branch of physics not yet recognised by mainstream science.

But none of that kills the gag.

Then Walter White arrives.

He punches Forrest Gump, watches the casualties quietly get up and leave, and walks directly towards the camera. He removes his hat and peers into the lens with the expression of a man who has just discovered that his Ring subscription does not include facial recognition for fictional drug lords.

Two years ago, AI video struggled to keep one person looking like the same person for six seconds. Now it can stage an 11-character slapstick routine, maintain the location and deliver a final punchline.

Hollywood has not been replaced.

But costume hire, casting, stunt coordination and the deposit on that hallway have all just received a fairly troubling demonstration.

And more than a few estate agents are starting to wonder how far they can push celebrities into their own property videos.

Keanu making katsu curry in the kitchen. Uma opening the bifold doors. The Joker looking genuinely impressed by the loft conversion.

For now, that would still count as misrepresentation.

By Christmas, it may be a premium portal upgrade.