- Categories:
- Softcore
- Masturbation
- Brunette
- Petite
- Big Tits
- Celebrities:
-
Hailee Steinfeld
Hailee Steinfeld333352810736
Full version (DFL): https://mrdeepfakes.com/video/77887/hailee-steinfeld-22-full-version-dfl
I thought I'd explore some single image face swapping options to see if it's finally possible to make decently high quality fakes with single image, while this one is not the best resolution, I must admit it did pretty good job for most part, even handled some tricky angles, still struggles with obstructions and masking is a bit shit at times but that could certainly be improved with a bit of DFL and upscaling, I'm certain it should be possible to even make half decent blowjobs with this but there are still some things it just can't handle, particularly major obstructions which throw off the detection, add to it the fact there is no way to apply that to extracted datasets means you just have to hope some shots come out and cut out all that doesn't.
For this comparison I did nothing with the results of the insightface one, no upscaling, no sharpening, nothing, just made sure all the clips had the same color grading (insightface result was a bit lower quality which altered colors a bit, my DFL deepfake was colorgraded so I brought the colors back to original, DFL deepfake has been post processed as I usually do (slight sharpening, standard color correction of the face to original face) but it's nothing major that improves quality in a major way, nor something that is hard or very time consuming so it shouldn't matter).
I thought I'd explore some single image face swapping options to see if it's finally possible to make decently high quality fakes with single image, while this one is not the best resolution, I must admit it did pretty good job for most part, even handled some tricky angles, still struggles with obstructions and masking is a bit shit at times but that could certainly be improved with a bit of DFL and upscaling, I'm certain it should be possible to even make half decent blowjobs with this but there are still some things it just can't handle, particularly major obstructions which throw off the detection, add to it the fact there is no way to apply that to extracted datasets means you just have to hope some shots come out and cut out all that doesn't.
For this comparison I did nothing with the results of the insightface one, no upscaling, no sharpening, nothing, just made sure all the clips had the same color grading (insightface result was a bit lower quality which altered colors a bit, my DFL deepfake was colorgraded so I brought the colors back to original, DFL deepfake has been post processed as I usually do (slight sharpening, standard color correction of the face to original face) but it's nothing major that improves quality in a major way, nor something that is hard or very time consuming so it shouldn't matter).
Unless you want better results, if you guys are interested I can do another test and pull out some tricks to improve quality, for that I can indeed make a tutorial.
Let me know if you want to see how well it can look with addition of better masking, upscaled and refined.
I don't like that they pulled the model from GitHub so that they could put it behind yet another shitty Discord bot.
You can get drastically different results if you use different picture.