Current status: Not saturated.
How granular are the concepts that AI video generation models have learned? They know "person", they know "fight". If they're gonna direct movies, though, they'll have to be able to take instructions that are much more specific.
Any white belt, such as myself, could execute (if sloppily) the below Brazilian Jiu Jitsu moves. Could AI, in the form of a video clip, show us how it's done? See for yourself!
Also, the scores are completely made up by me based on vibes.
| Model | Armbar | Triangle | Double Leg to Americana | Overall |
|---|
The following prompts were used for each move, identical across all models: