rerun119 I watched the "Threesome After Party 2" scene. It was OK. I definitely wouldn't say "we nailed it". The stroker is still jerky maybe it was worse before. It seems to me that sometimes the strokes need to be smoothed out. You hardly ever have a jerky stroke by a human.

Oh, it was way worse before. I rated those as 'not safe for use' 🙂
The current ones (at time of writing) Are 'safe for use' but 'would not pass quality check'.
This does not mean they are crap, just that we set a high bar.

Will65 CzechVR has been getting quite a bit of dissatisfaction from their scripts because they are limiting the speed and range of mothion to try to accomade every device

This is where our own made app(s) come in. They accommodate for this, instead of 'dumbing down' the actual script.

        That’s right. Back-end deployment took more than expected.
        Right now we are able to render 1 script in 20 min.
        We already greatly improved algorithms so all the scripts have to be re-rendered.
        The team is also working to scale 1000 scripts generation in an hour or something like that. So when we have another algorithm improvement everything will be re-rendered fast.

        It will take another month to set everything up the right way

        Rakly3 The current ones (at time of writing) Are 'safe for use' but 'would not pass quality check'.

        Thank you for the acknowedgement of this. My issue was crowing that "we nailed it" when it didn't seem like it would even pass quality control. That had me worried that this was the best we could expect.

          Glad to hear things are progressing with the AI. I am curious though how it handles non traditional scenes like gay, trans, lesbian strapon, handheld dildo, mmf, orgy, gangbang, etc., or scenes where penetration is outside the frame or otherwise obscured.

            doublevr hey so I wanted to try out the AI scripting on my pc, it’s a super computer of a machine so it should handle it no problem. I followed the instructions on SLR, but it seems to be totally off and doesn’t work as it’s making the strokes way to fast and they don’t seem to match the tempo. Is there a software I could run specifically to generate the strokes instead of uploading the ROI data tracking to the video?

                6 days later

                MrRadicalsheep Hi! What video did you watch exactly? What timestamp? Provide please more information, I will try to improve scripts.

                  2 months later

                  Here comes the first 1,000 AI scripts https://www.sexlikereal.com/tags/sex-toy-scripts-vr?type=ai

                  Little stops us from scripting most of the library at 1,500 scripts a day (might be 1,000 scripts an hour).

                  10 AI scripts downloads limit separate from 3 craft scripts download limit.

                  Passthrough, MosaicAI, automated Smartphone POV and many more come next.

                  Works on VR headsets and also smartphones now.

                    doublevr changed the title to 1,000 HighQuality AI scripts released - we nailed it 🔥 .

                    Very excited to browse through these alot of videos that no longer had scripts or videos I have loved for a long time that never had scripts are available now.

                    Tomorrow is Starfield and Funscript day for me !

                    I guess my other question is i am used to buying videos and scripts seperately to ensure I keep access even if the video is removed from the site. IS it safe to say purchasing a video with. an AI script ensures access to that video and script for the forseeable future ?

                    thatoculusdude

                    Funny that you'd ask about Lana Reif specifically.. I have actually been thinking about whether or not to script her french maid scene from DDF. Some time ago, I actually went as far as doing a quick first pass of the cowgirl part of the scene but just never found enough motivation to go and finish it.

                        What order are the AI scrips displayed? By script release date? Would be good to have more control over the sorting. Thanks. BTW if you have the data on which scenes a user has downloaded or favorited, you could have an option to show which of those scenes are now scripted, perhaps?

                        thatoculusdude Sure, I can put her in my working queue. I don't have a copy of the Dracula stepdaughter scene at the moment but I can grab it later. I've got to finish the current script I've been working on first. Can't say for sure when I'll actually get around to completing these Lana Reif ones, but I can definitely do them as time permits.

                            6 days later

                            Great stuff! I've been a bit skeptical of the AI scripts, but they have clearly improved and I'm so happy to see many awesome scenes scripted now, since I thought some of them would never get scripts lol.
                            I did encounter a couple of weird errors, but not too many. One recurring issue for me was that the movements were often too exaggerated. I do prefer long strokes to small ones, but I think it can break immersion when the movement in the video covers like 30% of the dick and the script almost makes a full stroke. But maybe this is done intentionally? I just don't think it's been that noticeable with hand made scripts.

                            Have you thought about integrating AI script generation into your manual scripting tool?
                            I played around a bit with the new AI scripts, and they are quite impressive. There are large segments where they are practically indistinguishable from manual scripts. But there are segments where they struggle (such as transitions between scenes)
                            I think there is an opportunity here to reduce the time manual scripts take by 5-10x.

                            You may consider using your model to fill in the high confidence segments, and leave the low confidence ones for manual scripting.
                            Does your AI model output a confidence score in addition to value that you could use as a cutoff? If not, can you potentially train a parallel classifier using the same feature values to predict confidence level?

                            • xoxo replied to this.