This was cross posted here because I'm not really sure where this belongs: http://preggophilia.com/deepfakes-not-mo...urbating-t
Morena Baccarin masturbating (Derived from Camryn and Dave XD)
Online Video:
https://www.erome.com/i/ckFMqmcQ
Same file as a download:
https://mega.nz/#!XHwGnKRL!0L4UOAgnDfkf3...qcbNM3EY10
This took about maybe 8 non-consecutive hours of hacking and maybe 5 days of processing on just my laptop CPU including many false starts. For example, Emma Watson was once of the first available celeb face sets. But Emma Watson doesn't really morph to Kinki Kristi at all. Pregnant star Erica J moves her head to much to get good face capture.
I have high hopes for Maise Williams (another easy to find face set) and Knocked Up Nikki. And I think that Pregnant Model Suki and Eve Myles may match well, but I'll need to harvest Torchwood for enough faces for Eve Myles unless someone beats me to it. Suki has a Angelina Jolie resemblance too and it wouldn't surprise me of a celeb face set of Jolie may just appear.
Unfortunately many of my favorite pregnant videos don't show the face stable long enough and have lots of profile and sideways face shots which deepfakes can't deal with that very well. So I'm really looking for videos where the face remains in frame, the face alignment doesn't change much and is more or less square to the camera, you don't get other faces in the fames, and has a resemblance to the celeb. And any hands (or other objects) blocking the face cause issues. Although the deepfakes-GAN codebase is already working on improving that. Frankly I expect the next year to do some amazing stuff in face swapping, even at all angles and the required similarity of faces will become less and less of any issue. You can see in my video when Camryn tilts her head too far back the face replacement stops working briefly.
I'm also finding that to figure out what is exactly what is going on with the math. I have to remember all this neural networks shit and gaussian math and marvel at all the linear algebra I forgot decades ago. But the help forum texts and especially the windows app FakeApp are making this all easier. So you don't really need to understand the math. This really is just me trying to figure out how it all actually works and a desire to take my dusty Gilbert Strang book off the shelf.
I'm not interested in taking requests. I have not owned a windows machine in decades so FakeApp isn't for me, even though it seems the easiest tool. And I've never bought a graphics card of any significant power. So I'm not taking advantage of all the hardware. I'm not sure that I'm into this enough that I want to blow $1000+ on a graphics card rig for this or pay by the hour to use GCP or AWS machine learning machines in the cloud.
I may figure out how to put this stuff into dockers and then I can spread them among the various cloud machines that I do have where pegging the CPU won't actually cost me money or subject me to scrutiny.
I encourage you to seek out the dpfak board if you are interested in trying this yourself. Especially if you have beefy NVidia gaming PC sitting around. Reddit closed the deepfake board and there is a clone on Voat. But Voat is really a bunch of raging toxic assholes, and I have a pretty goddamn high tolerance, so I'm sticking to dpfak for now.
I also predict that we will start to get models performing deliberate shows to make deepfakes source material. They can provide correct lighting, camera work, and even costuming to makes these fakes flawless. Winter Is Cumming.
Morena Baccarin masturbating (Derived from Camryn and Dave XD)
Online Video:
https://www.erome.com/i/ckFMqmcQ
Same file as a download:
https://mega.nz/#!XHwGnKRL!0L4UOAgnDfkf3...qcbNM3EY10
This took about maybe 8 non-consecutive hours of hacking and maybe 5 days of processing on just my laptop CPU including many false starts. For example, Emma Watson was once of the first available celeb face sets. But Emma Watson doesn't really morph to Kinki Kristi at all. Pregnant star Erica J moves her head to much to get good face capture.
I have high hopes for Maise Williams (another easy to find face set) and Knocked Up Nikki. And I think that Pregnant Model Suki and Eve Myles may match well, but I'll need to harvest Torchwood for enough faces for Eve Myles unless someone beats me to it. Suki has a Angelina Jolie resemblance too and it wouldn't surprise me of a celeb face set of Jolie may just appear.
Unfortunately many of my favorite pregnant videos don't show the face stable long enough and have lots of profile and sideways face shots which deepfakes can't deal with that very well. So I'm really looking for videos where the face remains in frame, the face alignment doesn't change much and is more or less square to the camera, you don't get other faces in the fames, and has a resemblance to the celeb. And any hands (or other objects) blocking the face cause issues. Although the deepfakes-GAN codebase is already working on improving that. Frankly I expect the next year to do some amazing stuff in face swapping, even at all angles and the required similarity of faces will become less and less of any issue. You can see in my video when Camryn tilts her head too far back the face replacement stops working briefly.
I'm also finding that to figure out what is exactly what is going on with the math. I have to remember all this neural networks shit and gaussian math and marvel at all the linear algebra I forgot decades ago. But the help forum texts and especially the windows app FakeApp are making this all easier. So you don't really need to understand the math. This really is just me trying to figure out how it all actually works and a desire to take my dusty Gilbert Strang book off the shelf.
I'm not interested in taking requests. I have not owned a windows machine in decades so FakeApp isn't for me, even though it seems the easiest tool. And I've never bought a graphics card of any significant power. So I'm not taking advantage of all the hardware. I'm not sure that I'm into this enough that I want to blow $1000+ on a graphics card rig for this or pay by the hour to use GCP or AWS machine learning machines in the cloud.
I may figure out how to put this stuff into dockers and then I can spread them among the various cloud machines that I do have where pegging the CPU won't actually cost me money or subject me to scrutiny.
I encourage you to seek out the dpfak board if you are interested in trying this yourself. Especially if you have beefy NVidia gaming PC sitting around. Reddit closed the deepfake board and there is a clone on Voat. But Voat is really a bunch of raging toxic assholes, and I have a pretty goddamn high tolerance, so I'm sticking to dpfak for now.
I also predict that we will start to get models performing deliberate shows to make deepfakes source material. They can provide correct lighting, camera work, and even costuming to makes these fakes flawless. Winter Is Cumming.
