But what will it cost? I would love to see this happen with the intent of making the end result free to use or very very inexpensive. Like 5 dollars a month. Then just get ad revenue. If I had the AWS or machine learning knowledge or a starting point I would put more time into researching it. But once the AI has been “taught” if it were my creation I would charge that monthly fee and let users upload a video a week and get a video/script combination spit back out of the machine.
Something along those line is what I had in mind and hope comes of this.
What I anticipate however is that it will only be used to make video/script combos for whatever respective studio can afford to create/patent the tool/concept…
It kinda is a bummer. I swear it’s something with the porn industry, because they always find a way to get more money out of the content they provide - either by creating a hundread different sub-sites or by paywalling other features like scripts.
With SLR “cheap” montlhy subscription at $20 you have to annually pay $240, add a few scripts a month to that at $4.00 average and things go a little crazy.
Lets not bicker like children but I will use this as an example.
If I’m paying for a service, I do not necessarily mind paying more, long as I’m getting more.
However if I’m faced with two difference services and one cost $45/mo while the other cost $25/mo but with additional charges $2-$5 charges, I’m going to pick the $45 every time because I know what I’m signing up for.
IMO SLR isn’t very clear that premium doesn’t include the scripts for someone new to the site. So a new people signs up, notices they also have to pay for the scripts in addition to the premium and they end up not using the server as much as they would if it was all included with the initial charge.
As a business owner, I think an all inclusive monthly subscription would yield more revenue then additional charges per video.
I will admit that micro transactions have proven to work well with video games but the porn industry is notorious for paywalls and I think it turns customers off.
However SLR owner probably know best, based on the info they that have from their subs. This is just my 2 cents.
Good points. Getting into interactives was super experimental for us. We were getting numerous requests, but we didn’t know where do we start - no scripts, uncooperative sex toys manufacturers, no documentation, no nothing. We somehow managed to get things work and everyone was fascinated how good it is.
It feels like it’s just now we are actually starting doing things. There’s a ton of work required to get the whole thing work and we are on the top of it.
There will be a ton more of development coming. Stay tuned
I’m excited to see the result/progress of this. Will you release an early build here for free when it comes to it?
I don’t think the time spent on scripts will differ that much tbh, but maybe the model training could help with atleast 50% of the time, and then you’d have to manual adjust the other parts.
Dropping my 2 cents here, having had some experience with training DL models and having watched a lot of comma.ai streams (George and the team drop some gold on there sometimes).
This problem won’t be easily (or even not so easily) solved unless a lot of resources are poured in (a potential workaround discussed later in this post). There are a few reasons I could think why this might be the case:
The sheer variability of the number of scripters, which in turn leads to even more variability in the style of scripts produced. (The same video could be scripted very differently, at least from a machine’s perspective by the same person in a different state). This needs to be smartly accounted for.
Since, the bounding box/masking approach (masking out genital areas or hands or whatever) speaking broadly is almost always going to be a limited approach (essentially you are cropping parts of movement that seem to be of little to no use), although it saves up on bandwidth/resources, it may not be the most reliable approach for the longer term solution.
This is a quote from one of comma.ai’s chief AI engineers (worded differently), any vision related problem would be best solved end-to-end, meaning you don’t teach a human how to drive by giving it a picture which has labeled annotations of what is what and where (it is an endless permutation of situations). An example of this in the context of solving the scripting problem can be, that funscripters take note of other cues (often subtle) apart from just the bodily movement while scripting, audio can be an example of this, or a part of movement that wasn’t deemed to be of importance by the masking tool.
I do not just want to be that guy that who says this is not gonna work, without actually proposing something that might. And I think what might work at solving a problem like this, is the following: I know there are a few galaxy brains on this forum, who are actively working on solving this, do us all a favor, and open-source the components you have already built. To whomsoever is the creator/mod of the forum, we can probably have a dedicated section for this discussion, where people can share the tools that they have built. There might initially be a lot of LQ stuff, but in longer-term it should lead to a better result for all of us.
The silver-lining that I realize while proof-reading this post is that even if we don’t solve this problem really precisely, it won’t be the end of the world, as in the end it is just simulated sex that we are discussing here and not some really critical issue.
Hey! I was actually the author of the idea gladly I have realized that it cannot be turned into classification problem from the data that we have (that came after evaluating it in a first place). So you are right about number data problem.
My plan is to add classification network on top of that. We will detect hands, mouths, feet etc and track them across the video (either tracking by detecting or tracking with bbox tracker). That should yield some reliable signals.
Again that won’t cover the 100% variety due to the occlusion, but I think we might get into some territory where it actually don’t whole lot of work for scripting.
If you willing to chat more just hmu philip at sexlikereal.com
I guess we will always have hand made scripts…
