And it’s a type of video that’s widely circulated online because of the prevalence of footage. I would be totally unsurprised if commercial GPTs have been given users’ private doorbell cam footage to train on, but I don’t think the fact they can replicate it well is compelling evidence on its own.
(Admittedly I don’t use genAI and don’t run into video slop very often, so all I can do is take the claim that commercial GPTs are “so good” at generating these at face value.)
Low res, fixed POV, wide angle lens.
And it’s a type of video that’s widely circulated online because of the prevalence of footage. I would be totally unsurprised if commercial GPTs have been given users’ private doorbell cam footage to train on, but I don’t think the fact they can replicate it well is compelling evidence on its own.
(Admittedly I don’t use genAI and don’t run into video slop very often, so all I can do is take the claim that commercial GPTs are “so good” at generating these at face value.)
generating videos is usually a diffusion model thing, not a transformer?
Not only are these not mutually exclusive, but Sora (as the most prominent example) is a diffusion transformer.
?!?!?!