...been bigging up the VAM/AI thing for a bit now, was not prepared for how good this one was:
https://www.sexlikereal.com/scenes/showcase-3-28544
Specifically, 14:18-17:42 (Blond spiky hair girl) if you want a single example from that reel. But they're all good, to be fair.
The script is absolutely on point, had to double-check at one point it wasn't handcrafted. What's most interesting is how little the AI actually has to work with, as much of the action is relatively deep, yet the AI was able to draw inference from the video, and suggested its own interpretation. As a human would do, instead of just blindly plotting exact [up/down] the script is made more suggestive, adding emphasis when needed. i was genuinely very, very impressed.
It's not 'all there' quite yet, i'd say the script is 98% finished, and lacks some of the clever tricks a human would use (like vibrations, for example), but it's actually getting there. Also, the simple front facing format of this POV video seems to help, lots for the AI to recognize.
But it's definitely in the right ballpark. DoubleVR was talking about community adjusted scripts, which is a brilliant idea, but i'm now wondering if you can't just program the AI to render out different interpretations of the same video as well (soft / hard / full / brief / edge / tease / demanding / wtvr) and run with those. Peeps can then cut and paste bits from different AI scripts if they want, to focus even more on the elements they enjoy. And then feed those scripts back into the AI, for even more machine based learning fun.
Nice... more SplineVR AI please...!
