FamilyDev
patreon
News! The new update will be released on 21.06!
🕑 Added 2023-06-16 17:36:32 +0000 UTCThe important thing is that all videos will have naughty sound!!! (patreon version only)
I also switched from remaking GIFs to remaking video fragments. It takes most of the time, but it gives the best result! It turns out that I use an extensive amount of the latest technologies, which is incredibly interesting. Today I experimented with deepfake. I was remaking a video with female masturbation, and there was a bad result due to the fact that only the face had to be redrawn (the overall picture is redrawn very weakly so that the fingers behave realistically and so that there is no original picture), the neuron redraws the face or not enough, in simple words deformed, and it already looks weakly like a face, or redrawn too so much so that the face was out of style, body position, lighting. It turned out that if, before redrawing each frame with a neural network, you simply layer the face with a deepfake, then due to the fact that the shape of the face is already correct, the neural network does not deform it. This is a completely new technique, which is showing itself very well so far. The only thing is that it cannot be used in blowjobs, since the face will be layered on the penis, I checked)
The time spent has slightly increased, but I will try to optimize it so as not to give a very small amount of content.
I am a pedantic person and I like to count everything and I hope it will be interesting to you.
Time spent on one 18+ scene:
1. Find a really hot scene that reflects the script and fits the entourage. 20min - ∞
2. Cutting the video into fragments that are suitable for redrawing (the face does not overlap anything, camera stabilization) 15min max
3. Structure these fragments and deepfake them to the desired face shape. 3-5 min for each
4. Redraw the frames of each scene with a neural network: the video is cut into frames and only the keyfremes are selected from them (those in which something has really changed, not afk). for each scene, individual settings are set and the redrawing is started. After the video frames have been redrawn, the frame review goes on. Bad frames are either redrawn by the neural network individually or I use my magic photoshop skills) It depends a lot on the GPU, I have 15 minutes for 1 second of video.
5. Another Ai completes the intermediate frames of keyframes. Then the result is made from the sequence. It depends a lot on GPU + CPU, it takes me ~20 minutes for 500 frames.
6. Optional. If the original video was of small resolution, then I use upscaler. Another 5 minutes.
I did the last 18+ scene of the next update for a total of ~20 hours, more time was obviously spent on redrawing frames)
