Scripting - Artificial Intelligence

General
21

Pretty confused about your post (I literally dont understand what you mean or want).

In my case, for example, I have a 25 minute video, with 18 minute being cowgirl and riding with mostly easy parts to tag and let AI do its thing. 18 minute of AI scripting takes about 5-6 minutes if those scenes run without any scene breaks (black scene, where the AI scripting stops and I have to start it again). Afterwards I check the frames quickly (around 30 minutes for those 18 minutes, sometimes less, sometimes a bit more) and tweak it quickly with some keybinds.

Then I get to script the rest of the scene manually and add some things here and there which can take a lot longer (1 hour +) or if the scene is easy to work with and the woman doesn’t slop the dick like it’s an ice cream I am quicker.

Vibration, sucction and tightness depend a lot on the sleeve the person has (and how sensitivy they are, vibrations are great on some days and on some others I dont even feel them and am more annoyed).

1 Like
22

Just stating that its the 80/20 ratio where the last 20% takes 80% of the time. Showing how effective (and useful) AI is on that.

But of that 20%, its mostly because devices cannot do all actions properly. If they could, AI most likely would have generated those as well. Lower accuracy then wont be noticed as the other effect compensates.

And long scripts do take significantly longer, but that is because the base time of testing such script (which is part of making it) also increases with it. A 10min video can take just 20min of test time (2 runs), while an hour can take 4h easily as its the entire video that has to work in relation to each other (first 10mins and last 10mins need to relatively match to each other). But even here, its the finetuning that takes most time.

But its that lacking of features where most of the time the extra effort is being done. And tools to optimize that (multiple script recordings, and a way to merge them by alternating script sources, and potentialy even use a channel as modifier - like converting vibrations) would be welcome here

23

I just have a quick question. Most ai generators i have seen try to mimic the motion of the “actors”. like selecting the hand/face/butt and so on. But would’nt it be possible to have AI recognice the penis and just track motions on that specific part. (like it recognises the penis look and length, and then tracks when something is obscuring it). not sure how to explain with out the use of images, and sure this would probably be best for maybe VR videos with a fixed camera angle. And it needs to understand penis motion though. But i have seen some tests where the entire female body is selected, and you chose what to track. Just a thought.

1 Like
24

Its not as simple as you are asking. As camera and actor position matter.

When both are very clear, the penis usualy also is, and the script matches decently regardless of movement. A POV camera is on that even easier.

But once the penis isnt well visible, it has to guess, and for that it tries to find reference points, points that regardless of camera position matches. AI can be made good to find out those reference points as even when something is obscured, the position for the penis doesnt change a lot. Its a thing AI does a lot faster, as machine learning code is designed to do this for you.

Thats why the penis isnt what is interesting to track for AI, its the entire bodies since regardless of camera, they should be visible.

Tracking the penis is on that even worse since it doesnt catch edge cases well. Penis length can prevent full insertion, yet sometimes position decides this. By using the body as reference point, regardless of position, a certain min and max distance will be detected (while still being considered as ‘being inside’). These values can be used to scale to a 0-100 distance. And therefor even nuances of not fully thrusting can be shown.

Sex rarely is a full 0-100 repeatedly. More often its only partial thrusting. Body tracking is very efficient here.

An example where tracking dick depth doesnt work:

  • in doggy/cowgirl style, one side dick is generaly fully inserted, yet the other isnt. For example in doggy, the bottom of the dick has a longer ‘travel’ distance when it would be fully inserted, but gets blocked as te top already is fully in.
  • in missionary, the dick is generaly going to be far more inserted. its a natural postition in which both top and bottom balance out. this makes strokes look a lot deeper. (the full insertion capabilities is 1 of the reasons why this is still one of the more intense positions - note most dicks arent long enough for doggy to reach the end)

Body position will be aware of this diffirence, and by using body distance will be relatively accurate at guessing pussy and dick location.

2 Likes
25

I guess that’s really what I mean. I don’t doubt that Google, Microsoft, or [Insert tech company here] could solve the automated scripting problem in a few months if they decided to. However, SLR is probably the biggest name in the game right now and I don’t think they have the resources to truly beat human scripts any time soon. That said, I would LOVE to be proven wrong. I hope that this comment ages like milk and you all get to laugh at how wrong I was.

26

Most of the time these things are more about costs. Its well known that projects consist of a triangle:
Costs, Speed, Quality

Costs for big companies is often something that isnt going to be a big factor if they realy desire an outcome. While its normaly where they do take a lot of cuts, its also the place that gets increased when speed is desired.

Speed is on that also not that much of an issue, since months is actualy quite a lot of time when trying to create something as trivial as just a linear motion script.

So in the end, its realy down to quality, and how far you want to stretch it. Its very easy to make a 90% quality script, and as a lot of information exists, those companies can decide to target towards 94%. This 4% might not sound like a lot, but that is because quality isnt a linear thing to measure when relating it to csots.

For its cost relation its usualy about the missing percentage, so 95% is actualy 100% better than 90% as the quality is doubled (you cant exceed 100%, but 100% itself is unreachable). And this translates linear into a cost rate to quality rate. Effort gets converted quite linearly. So spending double for double the quality, only adds this 5% onto the 90%. Making it feel like its only 5% of an improvement.

But spend a lot of money, and you can get that quality up. And note, computing power is on that a cost! A script can generate in a minute, or a day, and those scrips can have severely diffirent quality. Which with a good algorythm might become even better. Sure, its a repeating costs unlike a single development state. But that is because in AI the algorythms are often adjuted towards the target device that has to use them. AI is useless if the responses are too late. And if too fast, that gives margin for improvement by using more computing power.

But this also requires skill. Skill is a cost (people cost money). And people arent always available to buy, and can restrict those companies. It might on that cause a restriction on the desired quality, in such way they actualy cant beat a dedicated company.

1 Like
27

Does that mean you’ll have no use for scripters soon or is it just going to be a tool for us to create more?

28

Just a shame that society is so prude. Imagine if it was totally socially acceptable for companies like google, meta, or unreal engine to invest in adult tech instead of trying to block it out.

I also believe AI could be used to create very good scripts just as long as enough money and talent is available. Pretty sure a large company could even take things much further than two dimensional scripts. But hey, that’s not the world we live in. Still, advances in AI will keep happening and the community and companies like SLR will try to incorporate them in a NSFW context. So it’s just a matter of time.

2 Likes
29

First one looks like Rachel Starr’s ass, am I correct? It’s the only heartquake I like!

1 Like
30

Good eye! It’s a gif from this script.

1 Like
31

Nice, thank you for the confirmation.

Btw, have you found a way to create up and down points by frequency or speed? I’m sure there’s plenty of techniques to create patterns and strokes like the ones you did there but I don’t have time to look for the documentation of all the plugins and other stuff.

Actually I’m using Funscript Generator (motion tracking) and it’s good enough, I tried FunscriptToolBox MotionVectors with poor results.

Scripting is so niche that I can’t find good video tutorials that cover everything from start to finish.

32

This might be what you’re looking for. It’s what I use for vibrations. For that particular scene, I manually scripted the top/bottom of each stroke and then adjusted the frequency/amplitude until I got what I needed.

1 Like
33

Not really. Those are two separate things - mass market automated scripts and Premium craft stuff.

34

No one here is to beat human.

Automated scripts are always going next to Craft Premium scripts. It’s just a nature of things that is different. Like smartphone photos didn’t replace paintings, yet one can say those photos are much more popular than ever.

35

Note though, unlike humans, AI is better at certain tasks, especialy when precision is needed. Sure, our current quality of AI’s is lacking on that part, but in theory that limitation doesnt need to be there, so whatever you can see, the AI also can. At that point the AI can use a lot more reference points.

Humans then only provide a personal style to the video. And that is something that AIs eventualy also could copy.

These days you can see quite a bit of AI artwork showing up, and in many cases they beat human artwork already. And that is in both normal and abstract art. A smartphone photo didnt replace paintings because its the wrong example. A painting in this case must be compared to a photoshop drawing artwork. And while there is no 3d feeling visual effect from printing, this is generaly more a limitation for the printer than for the photoshop designer. 3d information is already storable, it simply isnt used as its not considered relevant.

The comparison is more like a CD and vinyl music. Where vinyl while being analog can in theory store more information, but is physicly restricted, and therefor requires some adjustments. And CD is only limited by its storage format, which while being digital can still store analog information. The format on that only being limited to a desired bitrate (in theory we can take it to 256khz and 128 bits, but no human will be able to hear the diffirence to 44khz/16bit anyway. studios on that only use higher values because it makes it easier to adjust things without instantly causing issues with compression).
Now consider that an AI can work at that 44khz/16bit level directly, while humans at best can only work with just basic beats. AI will have a lot more control on the music it makes. But AI rarely is used here, because its simply not needed, since for music its often not just the music itself, but band history etc and experiences that make the music the way it is. AI is in this case only used to enable a dead singer to be revived for a song. And not just that, even if the AI would work at that bitrate directly, the human ears cannot undestand the detail.

36

Do you think the offer to trade an SLR Original script for 500$ will persist, then ?
I’ve been heavily considering scripting for you guys full time so I want to see if that’s still going to be an option for at least a little while haha.

37

for sure :+1:

1 Like
38
39

These are AI script made on first pass SLR ScriptAI generated scripts are live! - #31 by Kieshi

Some really great feedback so far. Takes like 20min to get one. Working to get the whole database scripted