Does anyone believe that AI could harness the capacity to script something as good if not better than a human being if properly developed? I haven’t tried motion tracking but what for experienced scripters would you say is the percentage of precision it can generate?
Yes but I think we’d need a lot of testing in order to figure out what is “best” so that the AI can train towards it.
I’d say the motion tracking can script cowgirl scenes and some simple blowjob scenes pretty well, but it’s the more delicate scenes that it can’t do. I’m kind of a simple guy that likes the up and down strokes more than the extra details, but the “bottoming out” stuff is always a personal touch people add.
That little bit of style each scripted adds really makes it interesting. Some motions I would have never thought of sometimes fit so perfectly. I hope one day there’s just AI that can give us scripts instantly with little margin of error in its accuracy of what’s on screen. Thanks for your response
AI will never be able to do great patterns like Bekscript or 99Dm
We are releasing a second iteration of SexLikeReal automated scripts soon SLR Scripts AI Preview: AI Assisted Funscript Creation - SLR VR porn videos and interactive sex toys hangout
The whole slr database is expected to be scripted in July
Also we welcome everyone who works in the field to join our effort firstname.lastname@example.org
‘Does anyone believe that AI could harness the capacity to script something as good if not better than a human being if properly developed?’
…yes, absolutely, eventually. But there’s a lot of caveats as to how long that will take, mostly depending on what the sexual act you’re attempting to describe is, and how it’s encoded visually.
Where i believe the most interesting future marker to prepare for will be, is the point where the AI and the humans generate enough feedback between each other to birth the next quantum leap in scripting techniques, and in turn an entirely new generation of machines to script for…
It will be as good as a mediocre human, but not as good than a good one. And it couldn’t be better, because current AI doesn’t add anything, so it can never be better than what it was trained on.
I’ve tried automatic scripting and unless it is the most simple scene, with very clear one dimensional movement, no camera movement and the people in the scene are very static and in focus, it is super faulty. Any change of light? now it doesn’t recognize the objects it is supposed to track. If the penis is completely obstructed (because it is fully inserted, for example!) it freaks out and goes for the closest thing of the same color that is close to it, so from now on you are stuck tracking the movement of the leg.
Even the best results require an amount of human tuning that makes them not really that much faster or efficient than just scripting them yourself. Plus, as any skilled scripter can confirm, correctly tracking the movements in a script doesn’t equate a good script. The AI has no way to know when to add little extra touches or why are needed.
AI isn’t there yet, but I’d rather have a script thats 80% of the way there than one that doesn’t exist at all. I’m optimistic
Loving the feedback
Tuning a badly done script (considering the time it takes for your computer to do it altogether for each time you re-run the auto-scripter!) takes more time than doing the script well from zero.
I’d rather have no script than a bad one, by experience. Specially if I have to config the machine to auto-script it too.
In terms of accurate, true-to-life range of motion, yes. I am confident that AI will be able to detect precisely how far one body part moves from another and spit out a number with a greater degree of precision than a human. I also believe that it will be able to do this more consistently and FAR faster than a human.
I think AI will make an absolute shitload of very good, but not GREAT scripts. I will 100% download and enjoy these scripts with minimal complaints. Seriously, I hope AI makes my scripts obsolete!
However, there is a certain ART to scripting that I’m not sure AI will ever truly master. This is my greatest achievement as a scripter. Sure, I COULD have just treated this motion as a straight up and down line, BUT that wouldn’t accurately capture the fucking earthquake going on here! I had to make a creative call to create the skeleton up/down doggy action and then fill it with vibrations. I also had to determine the frequency of those vibrations in attempt to match the shaking as best as possible.
I don’t think it’s IMPOSSIBLE for AI to learn that level of complex scripting, but I think it’s going to be a while before we have AI that’s “creative” enough to make those decisions.
This depends on angle, speed, motion complexity (see above), and video quality. Something like this?
About 80% accuracy. The height of the strokes is almost always off and the timing can get a bit wonky.
I see what you’re saying
80% is pretty damn good for only 20% left to tweak, + bells and whistles
For sure! I automate simpler stuff all the time and then tweak accordingly. My only point is that I think it’ll be a long time before we see truly great AI scripts.
AI can definitely stomp out script that are for animations. alot of animations are a simple loop or if the animators like to suffer for their art it will be much more detailed. Still tho it can do animation quite well vs IRL porn which definitely requires a human touch to make it work.
Pareto’s Principle: that 20% left is the 80% of the work anyway.
What the AI can do because it is very simple is as simple to just do by yourself on the fly. That 20% is where the real work is.
I think the biggest change over time is going to be the devices themselve. This shaking effect might simply become a vibrator in the device. And just like suction it becomes a seperated feature.
And note, this shaking can be done in several ways:
- Basic vibrator
- Vibrator using a more ‘impulse’ based system (instead of rotations, it will have a weight move to force this shake)
- Rotating beads or similar system. Instead of it being vibrations, its an actual object pushing the sleeve. (like for example: Milk Me Silly – PDXBrands )
AI in most cases for making scripts will just use pattern recognition. And just its detection has to improve.
For internal movement, AI can do some predictions as for example tightning might be harder to guess. But instead of it guessing randomly, it just uses other reference poitns (womens face, audio, video position).
The real problem is simply: it doesnt matter as the devices themselve are simply not complex enough to realy get the advantages. Why try to imitate these rare cases that would be trivial with an other system?
Doing those simple things on the fly takes me 5x longer than doing it through AI and tweaking it. Fast processing is a godsend when it comes to speed. E.g. 10 minute cowgirl scene takes about 2-3 minutes for my computer to process and script and then tweaking is a matter of cruising through the points and making sure it works well. I agree for the more difficult things which you cannot automate and have to script manually.
I see it slightly differently. I think AI could absolutely knock these out to a 99% accuracy, including the subtleties you are showing in our best manmade scripts, but that takes significant data and model training and the AI industry is focusing on other things. If Google wanted to take a crack at AI funscripts, they would nail it. Until then, we are all we got lol. AI can work through millions of iterations and work with well over thousands of spacial geometry points in a few hours. It WILL take a long time before we see good AI scripts but not because of the limitations of the technology.
Most of the time its just device limitations you face. Those harder sections are often because either the device isnt fast enough, or because it wants to use a feature that the device doesnt support (vibration, suction, multiple hands, tightness).
Those harder sections are just workarounds, thats why they take that many time to make. If such things arent desired, AI can generate about 80% of the script at very high accuracy, 15% at a high enough accuracy it generaly doesnt even need a change. and 5% of the more difficult parts where AI is generaly not bad, but does require fixes.
And time wise, you will spend about 20% of the time on the 80% part (just watching it), and the remaining 20% is that part you adjust (which takes about 4x as long then).
But its not like trying to adjust for those is a must, it heavily depends on the video to begin with. If the movement is fast, suction isnt realy needed for the experience on the handy. Double action is on that generaly the most difficult since it doesnt allow a simple module added onto it to imitate it. Vibration, tightness and suction already exist on strokers (they just arent script automated), and often are too heavy. That is generaly an issue that will be solved over time (with a bit more expense as a result like the osr).
Idealy you even want the AI to generate a script for each of those things, and that includes hand/head movement as seperate scripts, and then have a script program to allow merging.