Hi all. This thread came about from a discussion last week. The folks at Sex Like Real (SLR) have presented their latest iteration of their AI scripting process and have asked for feedback.
In the spirit of full disclosure, I’m not an employee of SLR, but I did receive a weekend of free access in order to do this write up, so bear that in mind that I did receive consideration for this piece.
Despite being a craft scripter myself, I’m generally in favor of AI scripts and in the past, have been impressed with the SLR AI scripts. They’re not anywhere near perfect but once they get going, they’re accurate enough to get the job done. I see -a lot- of negative criticism about the SLR AI scripts here on ES, some of which I agree with, others I don’t. I’ll put some thoughts on that in my summary.
But in general, I think in it’s current state AI is capable of scripting clear, cleanly shot stretches of repeated action pretty accurately. Where it has issues is the “connective tissue” between these stretches of repeated action. By that I mean the slow initial entry at the beginning of a position where the actors are getting acclimated, grinding sections where the penis is obscured, and the furtive gestures that happen when a scene transitions between things. Here, the AI struggles to interpret the visuals to translate it into action.
Over the weekend I went through quite a number of the SLR AI scripts to spot check what they’re like. Additionally, SLR sent me two versions of scripts for the same scene so that I could analyze the improvements that have been made to their AI process. On the random AI scripts I checked, I feel like my previously stated opinion was reinforced - once things get going, they’re really very accurate but it’s the connecting bits between the long stretches of repeated action that are distracting.
The scene I took a deep dive in was Pump A Baby Inside... - VRHush - VR Porn Video | SexLikeReal, a gripping examination of one women’s quest to get knocked up n’ milky. Now, there are aspects of this scene that make it really difficult to script - lots of grinding and one extended sequence about 30 minutes in where the scene is cropped so that there is literally no action going on onscreen and shadows offer the only hint of what’s really going on. Maybe that’s why it was selected to have an AI script made as a proof of concept.
The two scripts I was provided for this were both done by AI and not edited by human hands - that’s what SLR told me and I certainly found no evidence of any editing to correct obvious issues. Essentially, they tweaked their AI processes and produced a newer script. The differences are pretty subtle - in the lengthy, repeated action sequences, the two scripts are virtually the same, there are some imperceptible changes to height and very rare changes to the points of articulation. The parts between are where we see the changes, which are still fairly subtle.
This is a grinding section early in the scene, the new script is on the top, the old on the bottom. Here we see where the AI captures more of the grinding that’s going on than before, you can also see later on in the timeline where the AI makes adjustments to the intensity of the stroke to better match the action.
Next we look at a bit from a cowgirl section where the angle is such that you can’t see the in and out. These bits are tough to script and you can see on the bottom where the original version of the script has a pretty intense range of motion where the newer iteration is more toned down and realistic.
Here’s a section where it’s cropped so you can’t see any action onscreen. These parts are really unpleasant to deal with as a scripter. The best you can hope for is to create something that doesn’t ruin immersion. In the older version, you can see where the AI chokes a bit and leaves it flat. In the newer script, the AI uses strokes running almost the entire length of the device. This is one example where I feel like the original was a better way to treat that section as the new script definitely pulls me out of immersion in that segment.
Here’s another bit from the same segment. In the older script, the AI seems to be trying to match her action as she plays with herself - hence the jagged red line going from bottom to top. This is likely due to the AI trying desperately to lock unto what the fuck is going on. In the newer script, the AI maintains a less distracting line.
A third bit from the same segment - here the new AI script is definitely an improvement as it catches pretty accurate movement whereas before it kept it pretty flat despite being able to see motion in her body as it reacts to being fucked.
Towards the end of the scene, here’s a quick bit from some of that “connective tissue” where she’s using her hand as the scene transitions to the finale which is a vigorous cowgirl part. Here, I feel like neither script accurately represents the action. It’s a quick bit that’s over in a second, maybe it distracts you, maybe not, but I think it illustrates my point about these bits being where the AI is weakest.
As you can see, it’s a bit of a mixed bag. Overall, it represents an improvement however and I think that’s how this thing will progress - in small steps rather than huge improvements.
In summary, I think SLR continues to be on the right track in improving their AI scripting. I understand some of the criticism leveled at SLR here but I think a lot of folks fail to consider the fundamental differences between what we do here on EroScripts and a for profit entity like SLR which has a radically different approach and business model. EroScripts is an ala carte site. It takes some practice to find what you’re looking for but once you do, there’s a wealth of great, free scripts to try. But it’s very “scene” based, you download one scene and you want a script that is super accurate in order to watch that scene from top to bottom.
While SLR does allow for ala carte purchasing of scripts, their main focus is to create a subscription environment. I suspect their ideal user is someone who browses their content, watches a bit of one scene, moves onto another as their whims change until the user is “finished” being entertained. From that standpoint, I think the AI scripts definitely add to this and add some level of interactivity to a greater selection of scenes on the site, thus adding value to the subscriber. Buying some of these AI scripts for $4 a pop? Yeah, I totally understand the concern about this. There’s at least two scenes, likely a bunch more, where scripts are offered here for free that are far better than the AI scripts being sold by SLR. But SLR is trying to make as much of their collection interactive as possible and they have a massive collection of stuff on there. Again, this benefits the subscriber model where you have users “dropping the needle” on a bunch of things to see what they like. But yeah, I still have not found an AI script on there I’d be happy if I paid more than $1 for buying it ala carte like that.
Anyway, thanks for reading, sorry so long.