• Content
  • Passthrough AI Mask generations and improvements

Vrsumo2017 I really like that missionary angle, it's similar to the one Wankz uses.

It's great because it's still plenty close to her face, but with her legs back you can see everything and it feels more intimate with her legs up around your torso.

    HughMungus yeah bro, and the scale be looking on point too, looks life size on my bed in mish and with a pillow behind they head, its crazy bro, and with a torso...it's the best!

    19 days later

    doublevr cool. Will be looking into each of these. Let's keep them coming 👍

    Words are free, action is desired....I am little bit disappointed cos there has been 3 and half months now since I listed these problems quoted below. Nothing has changed, nothing has improved 🙁 I have few more known issues, but why an Earth would somebody give you more examples and ideas how to improve your AI mask if you have done zero about it? You have had plenty of time. LLM-models and AI had improved a lot in this time, but what about SLR's AI mask? Nope. Such as shame.

    Picouser I am so confused. I hoped this was a joke but probably it was not 🙁

    I don't know how did you creat this AI mask in SLR, but if you have watched AI porn even a little bit you should have plenty of ideas for improvement. I don't want to be rude but shame on you. How can you build AI mask and not to have proper roadmap for improvements. It is like Tesla would build its first EV and then don't know what to do. Mind-boggling.

    Let me give you some video examples and starting times:

    https://www.sexlikereal.com/scenes/raw-taboo-peeping-43889?t=179
    Background is obviously seen between models body and arms (under armpits) and between legs. White carpet is also blinking next to hip. AI mask can not decide is it skirt or carpet.

    https://www.sexlikereal.com/scenes/hot-petite-stripper-rides-hard-cock-in-the-vip-room-43369?t=668
    Models black socks are transparent (on and off) because algorithm can not decide is it floor or sock.

    https://www.sexlikereal.com/scenes/hot-petite-stripper-rides-hard-cock-in-the-vip-room-43369?t=782
    Same video little bit further. Look at the male left hand and girls shoulder. AI mask edges are not smooth at all. If you can pause video perfectly you can see edges are like Lego blocks.

    https://www.sexlikereal.com/scenes/baby-its-cold-outside-7091?t=310
    Undressing the sweater it becomes transparent in front of the model so the model is also transparent. If you have fabric or other item touching the model it can't be transparent.

    https://www.sexlikereal.com/scenes/mirror-mirror-16617?t=21
    We don't need models reflections (glass or mirror) in AI PT. Mirror works great in normal videos but not in PT. It is just confusing.

    Summary
    Armpit and leg space problem
    Smooth edge problem
    Mirror problem
    Fabric and item problem

    I am little bit sad this tread did not get proper interest. AI PT is one of SLR competition advantages. Why not handle and improve it such a way also?

      bruce Why have you stopped generating AI PT for Wankz and MilfVR? Is there another 'trade war' going on like last time when they didn't upload new scenes for months? I will take silence as 'yes but once again we are not at liberty to talk about it in public' 😅

        I've asked this before but never got an answer, why can you only adjust 1 colour when masking out live cam girls? The 3 options are there but it's only possible to alter 1 colour, is this a bug or a limitation due to a live broadcast? Would love to know as the 3 options is so handy, a lot of times you need at least 2 as most videos (especially live cam girls) have more than 1 colour background.

          Picouser We are working on that problems to implement most of that issues into PT 2.0. Currently we preform put attention to the research on more temporally consistent mask generation, which lead to more stable and mask and largely improves problem with the edges and topology (hole filling). Here are current improvements (top is current AI PT, bottom is model which is in development currently)
          The problem with objects is the hardest to solve due to unclear logics of which objects should be marked as interactable, and not but we work on that too

            AlexSLRLabs would be really great to have better tracking of clothes, especially costumes