An LED volume is a curved wall of screens you put actors in front of, so the background is really there on the day and the light coming off it lands on their faces properly. The Mandalorian was shot this way, and plenty of television has followed.
Ours doesn't exist. We generated three versions of it: same framing, wall off, wall running a calibration pattern, wall playing a beach.
The voices, the pauses, the way the camera drifts in and the warm light that lands on us once the beach comes up are all model output. Nothing in the shot was filmed, and the room was never built.
Seedance 2.5
Seedance 2.5, ByteDance's latest video model, does thirty seconds in a single generation, up from fifteen in Seedance 2.0. You can give it up to thirty images, ten video clips and ten audio clips as reference in one pass, then pin instructions to specific timestamps.
The jump from fifteen seconds to thirty is what let us do the whole exchange in one generation. At fifteen we'd have got the setup or the punchline and had to cut between them.
The Raq.com test
We ran a harder version of this for Raq.com: a thirty-second take where Raq Raccoon walks through five rooms, each one a different Raq.com module.
Eight reference files: two character shots of the raccoon, five room concepts, one motion reference. Every room change is physical rather than a cut. A card slides past the lens, the camera walks through a doorway, a photograph swings across frame, and when it clears you're somewhere else.
The whole thing is written to the tenth of a second. The full prompt is on the Raq.com post if you want to see what that looks like.
Limitations
It still can't produce reliable text. We used a whole section of the raccoon prompt to keep the jumper logo legible and replace every card, control, label and clock face with abstract bars.
There's no public API yet, so you can't connect Seedance 2.5 to your own workflow. Until there is, send the brief through Raq.com Hybrid and we'll return a finished video within 48 hours. API access is coming to Raq.com AI Video Lab.