Comment by pointlessone
2 days ago
Some of the shots are impressive but… Even among these hand-picked examples there’s a plenty of unnatural movement. And it seems like it was trained on the most hyperactive subset of tiktok as it apparently can’t hold a scene for more than 5 seconds.
While it pulls some pretty difficult things, it seems to struggle with other *seemingly* simple ones.
The piano in the beginning or the photo camera used by the photographer has "AI text" written on it. The old man with the beret in the cafe goes through his beret with his hand. The girl on the seaside looking back turns her head too much almost like an owl. The boy-in-a-bike-through-an-ewuropean-city scene ends on a square with an amorphous being in a unicycle under the tree...
ByteDance has been testing their model on the Model Arena for weeks. They were covertly calling their model "Unicorn" until just a few days ago.
It's already ranking better than Google Veo 3:
https://artificialanalysis.ai/text-to-video/arena?tab=leader...
By a LOT too lol, not even close, wow, that said, i imagine if they enabled veo3's sound... it would win lol