• SLR

  • What’s the next top priority effort we should hop on?

xoxo

I'm on a Nvidia 3090 with i7-8700k so not the ideal hardware setup. I'm still optimising all the low hanging fruits I can tackle.

I will try and provide a benchmark test very soon for a 60 sec clip (single and multi-threaded). Bear in mind that the code is written in python so it can be optimised much further if you refactor it to a custom C++ CUDA kernel.

Anyway I'm happy to gift my solution to SLR and see what your computer vision / AI team can do with it.

But aside from the technicals, I believe this business model of ethical deepfaking with consent can open up tremendous amount of oppurtunities for onlyfans girls, cam girls and others who don't want to perform but have no issues with licensing their face for a royalty-fee

      damian oppurtunities for onlyfans girls, cam girls and others who don't want to perform but have no issues with licensing their face for a royalty-fee

      If i understand the concept right, you could for example merge a licensed face with the body of a performer in one of the videos owned by SLR. I get it why you might want to deepfake faces from outside adult entertainment into a porn video, but if it is about your favorite onlyfans or cam girl, what you will get is a chimera of that girl's face and Melody Marks' body. Even if the VR performer's body looks quite similar, you will probably not be able to ignore the fact that it is not the right body. Of course I might be missing something here, but for me ethical deepfaking sounds more like a gimmick for a few than like tremendous opportunities.

          SchnuppiLilac Well every great invention starts as a gimmick but gets polished over time.

          Personally I don't use OnlyFans so I can't really speak on body + face attachment a fan can have. But I totally get where you're coming from and it's valid criticism to this idea. If you've been exposed to the face and body too many times and then see a deepfake video where it's only 90% there, your brain might have a tough time to click. Sourcing the right VR performer with the right face takes time but there's some fun in that. Nowadays there is a large pool of videos to choose from so you can get super close (at least I have) where I bet most consumers wouldn't have an issue with.

          On the other hand I'm in the camp of "better something than absolutely nothing". This tech is moving at rapid speed and I'm sure by next year we'll be seeing StableDiffusion / LoRA / Controlnet models to transform entire bodies as well. Best is to already work on proof of concepts and add new modules to make it better and better to stay ahead of the disruption curve that is yet to come...

          By testing it out ideas in the market even with a 95% accuracy rate, you can see what the feedback is and attract talented devs who are willing to take it to the next level. Passthrough VR on the Quest 2 is a great example of that. It's a gimmick right now but it's naive and short-sighted to think this is not gonna become a large category of porn in the future.

              RealcumberScripts Being able to switch on the fly between pov and voyeur in the middle of the same scene would be pretty amazing and hopefully possible in the near future

              Yes, that would be awesome!

                RealcumberScripts You simply cant get those amazing voyeur body views and angles in POV that you can see in voyeur/non-POV (or even proper closeups alone)

                Being able to switch on the fly between pov and voyeur in the middle of the same scene would be pretty amazing and hopefully possible in the near future

                Great idea! Could be done right now by planting more camera's but mask them nicely within the decoration.

                Or hopefully we see some advancements in Nvidia NeRF technology which is quite exciting as well:
                https://developer.nvidia.com/blog/getting-started-with-nvidia-instant-nerfs/

                  @damian Stable diffusions and all this stuff is cool, of course. But to move this into production - another story.

                  To perform any body transformations or deepfakes or whatever we have two options:

                  • Prepare upfront and just stream the results (Scripts, PassThru)
                  • Do in real time (DeepFaces, Body/Face-tuning).

                  With first approach there are some technical limitations but we can overcome them.

                  With the second one there are much more problems:

                  • We need to infer NeuralNetworks on headset.
                  • Taking into account bitrate of our videos and headset hardware doing this becomes close to impossible at the moment.

                  We can turn second approach into first one and think about presets. But with DeepFaces this is kind of illegal I guess😉 With Face Tuning and similar things - unpractical.

                  Any ideas and ways to overcome this are welcome. We don't think too much about this at the moment.

                    I'm patiently waiting for flat haptics. Particularly the handy since its not on haptics connect yet.

                    More scripted content of varied genres and scenarios, which I guess AI might be able to do now that I think of it.

                    Y'all have been addressing this and bless you, I just want more lmfao. Most of the scripted content is pretty vanilla, which is totally cool and I'm usually looking for... Sometimes I wanna mix it up though 🤷‍♂️

                    It'd be easy to script a decent chunk of other content (at least the good scenes in different genres) and get it out of the way. Then you'd have more niche and variety on the site which I presume would be a good thing

                    BDSM, Orgies, MMF, Interracial, Gangbangs, etc.

                    Oktoberfuckfest was a great scene, the orgy scenario was hot... I think it's the only scripted content where it's an orgy like that.

                      Choose your own adventure scenes would be great. Right now every scene is just what it is and you get people complaining constantly about what is, and is not, included in them. It would be great if you could just film everything and when you went to play it on the app or the website there would be a "menu" where you could just pick what you want to see and the website would stitch that custom video together for you. Don't like anal? Don't choose it and you won't see it. Weird armpit fetishists could get that stuff and everyone else could just not see it.
                      You, or another studio, could have a virtual brothel series that would work well for this. Choose the girl and choose what you want.

                      xoxo Yea you're right, moving this into production is not that easy. The models are becoming quite powerful but still sometimes misses or misaligns when the face moves around to the sides or when throwing head to the back.

                      On the illegal aspect, I didn't mean allowing users to upload their own source face to deepfake. That's gonna be a liability hell on your end that I wouldn't wanna touch either. I was strictly talking about a licensed SLR Face Gallery to choose from with consent all curated by your team.

                      Oof never thought about the second idea but gonna take a while until hardware catches up... maybe with the Apple Vision Pro 3th gen lol. I'll keep exploring and report back if I find a breakthrough, for now I wouldn't make it a high priority

                        cowdoy we will be posting new AI scripts every week for different videos, you can check it out in this thread
                        You can also mention in that thread if you want any particular video to have an AI script, so the team can see it

                          Do you guys curate your content? Obviously, if you want to be the Netflix of VR Porn then you have to maintain a certain standard for your production partners. So scenes which are clearly, objectively not up to any kind of professional standard should not be on the site IMHO. I can't upload my home movies to Netflix and I hope you have some kind of quality threshold for your content. You should be watching everything you show, and watch it on an actual headset. If it's not real VR or is basically unwatchable then it shouldn't appear on SLR - IMHO. Just a thought.

                          Sire, please make passthrough on Virtual Desktop (with Oculus or Steam Version) possible. There's already an passthrough-environment ... so it can't be that complicated. If you have the guardian activated and move slowly out of it, it is like at "the first SLR - passthrough testing" ..... lol .... feels like 10 (2) years ago somehow 🙂

                            Sorry, I am not able to tell the single technical steps.
                            The thought was that the passthrough environment (that's there anyway there as a background / homescreen in Virtual Desktop) will be underlayed if you press the passthrough button in the DeoVR App.

                            doublevr the oculus desktop version of DEO has passthrough settings but passthrough doesn't work on Quest via virtual desktop yet (instead of seeing your environment you just see black). In the past, this was not possible because VD did not support passthrough. This recently changed - Virtual Desktop now has passthrough support so PCVR apps like DEO should be able to enable passthrough.

                              5 days later

                              doublevr Maybe it could also be an idea to use strong GPUs like the 4060 and below to supersample (or whatever) the 'crappy' Quest Pro/Pico4 passthrough? Or use some filter/decode technique....
                              Don't know if that's possible though (: