[3deroticer]

http://i.imgur.com/yJC7ckY.gifv

Found some deep fake of Emma Watson and decided to BE the animation

[Rin Tohsaka]

http://i.imgur.com/yJC7ckY.gifv

Found some deep fake of Emma Watson and decided to BE the animation

Ever since I found out about deepfakes around a year ago, I've always been disappointed that so extremely few people try to use it for things like breast expansion and what-not - people always just seem to focus completely on re-creating what already exists in real life...

(I once tried to make a deepfake of my own, by my desktop PC is so lacking in the graphics department that it would take a week or more just to make a very short deepfake that isn't of crap fidelity...not to mention it's my impression that the primary deepfake-creating software requires an Nvidia GPU for GPU acceleration, and I prefer AMD and/or Intel graphics due to their much friendlier Linux compatibility among other reasons)



Anyway, so as to actually provide some content of substance to this thread, here's an image I came across on DA very recently by an artist that I don't believe has been mentioned in this thread before:

[Megan]

Emma Watson Massive Udders Ready For Milking.

[T_Man]

 

[3deroticer]

http://i.imgur.com/hm6SFPN.gifv

Emma Watson deepfake video and then BE
I did the BE part!

[solvegas]

http://i.imgur.com/hm6SFPN.gifv

Emma Watson deepfake video and then BE
I did the BE part!

Damn good ! 

[Rin Tohsaka]

For extra fun and realism, I turned that animation into a higher resolution, higher framerate loopable video - it might look a bit "overly smoothed" in a sort of oil-painting kind of way though, and there may be a couple minor interpolation artifacts...

When the video is playing, right-click the video area and select "loop" to have the video loop.

Full credit to 3deroticer for the original animation
Most compatible (H.264): https://ttm.sh/hBJ.mp4
Best quality (AV1): https://ttm.sh/hBo.webm
1 step below best quality (AV1): https://ttm.sh/hBr.webm


I had to upload to a temporary file host (they should remain available for a few months) because, despite the video files being less than 3MB (the "best quality" may not be however, hence why I also uploaded a "1 step below best quality" which is definitely below 3MB), the forum wouldn't accept them...and by now it's much too late in the day for me to want to bother with re-encoding at higher bitrate and the like as I'm practically falling asleep over here.

So yeah, if someone is able to successfully attach these videos to an actual post on this forum, that'd be great.




I can't help but wonder how feasible it'd be to write software to take advantage of the sort of neural networks/machine learning stuff in a deepfake and/or DAIN-like manner to change the size of breasts in at least a semi-convincing manner...but it's PC hardware that's my expertise - I couldn't write a line of code to save my life.

http://i.imgur.com/hm6SFPN.gifv

I don't suppose you have a higher-quality / higher-resolution version of that exact same animation, whether as a native video file or something like an animated PNG?

Nevertheless, for anybody wanting the absolute maximum visual fidelity (which is a funny thing to say considering GIF's 256 color limit ), here's a link to the actual GIF file instead: https://i.imgur.com/hm6SFPN.gif

Oh and btw, the source video you edited seems to have telecine (every 4th-5th frame is repeated).  Now I've no idea how you actually make these sorts of edits, but I would think that it'd be ideal to remove any such repeating frames so as to reduce the quantity of how many frames need to actually be edited (in this case it would have been a quantity reduction of 20%).


----------------------------------------------------------------


For the technical stuff on what I did...

tl;dr:

• virtualdub2 to export GIF image to individual frames
• plain old windows explorer to remove telecine'd duplicate frames
• waifu2x snowshell to upscale and remove dithering
• xnview to batch rename a copy of the frames in reverse to facilitate looping
• avidemux to encode a lossless 24.00fps input file for svp
• svp4 (linux ver.) to interpolate to 60.00fps
• avidemux to encode the "most compatible" h.264-encoded video
• my mp4box gui to mux the h.264 video into an MP4 container
• ffmpeg nightly and this guide to encode the "best quality" AV1 video

Everything and the kitchen sink:
What I did is that I downloaded the actual GIF file (not the MP4 that imgur transcoded from said GIF), opened it in virtualdub2, exported the frames as an image sequence, removed every 4th frame (the video seemed to have been telecine'd, whereby every 4th-5th frame was repeated) by just having the images aligned in the windows explorer so there was 5 frames per row and just drag-selected the 4th column and deleted them all, used xnview's batch rename to have the file names increment by 1 (thereby making up for the gaps caused by the removed telecine'd frames), ran the frames through waifu2x snowshell configured to 2x scale (960x480) and noise reduction set to max (anything lower did not remove the dithering), made a copy of the waifu2x'd frames and deleted the first and last frames and used xnview's batch rename to "reverse" the frame filenames (so frame 201 was renamed to 203 while frame 001 was renamed to 403), added the renamed reversed frames to the normal waifu2x'd frames and opened all frames in avidemux where I exported as a lossless video set to a framerate of 24.00fps (after removing the telecine the framerate was like 26.66fps...which is no good for the next step), then took that 24fps lossless video and opened it in SVP4 Linux to interpolate to 60.00fps with maximum interpolation settings (cause I'm a sucker for high frame rate; the closer to an exact multiple the better, but 30fps looked too fast and 20fps too slow, so the 2.5x of 24->60fps was the 2nd best choice unless I wanted to interpolate to greater than 60fps...but I'm probably the only one on here with 75+Hz displays) and encode to ffhuffyuv lossless, and then open that ffhuffyuv lossless video in avidemux with a filter setting the frame rate to 60.00 and encoding to x264 configured to high profile + placebo + slow 1st pass + 1490kbps average bitrate and export to an MPEG TS container (exporting 60.00fps videos into an MP4 via avidemux can cause a framerate bug), then use my mp4boxgui" to and demux the h.264 video stream from the encoded .ts and then use my mp4boxgui to mux extracted h.264 stream into an MP4 container for the "most compatible" video file.  Then I followed this reddit guide, changing the "mp4" extension of the input file to "mkv" and the "30" to "29" to encode the SVP-encoded lossless 60.00fps video into AV1 for the "best quality" video file (I left it at 30 for the "1 step below best quality" video file).

Several of these steps could have been avoided via the use of command line, but I've no idea how to use x264 via command line.

Also I'm pretty darned sure one could achieve higher quality results with other software, but I was simply using freely available software that I already had installed (note that only the Linux version of SVP4 is free).  If I feel up to it, a higher-quality version could be something I may investigate into at some point in the future.

[3deroticer]

I won a year subscription to the latest creative cloud from Adobe lightwave contest. So I wonder if there is a filter to get rid of duplicate frame? My feeling is that the more often you convert a video its starts to degrade, that is why I didn't get rid of the extra frame. Its just quicker not doing that. When you realize that I have to rotoscope a mask on every frame for the BE effect on each breast, you will realize that its already super tedious of a task. I also have to create a separate video for each breast before doing that. So already I degraded the video at that point.

[Rin Tohsaka]

My feeling is that the more often you convert a video its starts to degrade

I have no idea how the software you use works, but for digital formats it all comes down to lossless vs lossy.  The easiest way to explain this would be with other, more well-known formats, such as with images.

• BMP and PNG are lossless - if you save, edit, and resave over and over, no quality will be lost.
• JPEG is lossy - if you save, edit, and resave over and over, quality will be lost.
  
• GIF is kind of a hybrid - it's lossless up to 256 colors, but anything above that becomes lossy.  Luckily animated PNG (aka "APNG") is a thing now which, like PNG, is truly lossless up to some billion colors (16bit per channel including alpha transparency) and, as the name suggests, supports animation just like GIF does; it took a while but APNG's browser support is finally quite widespread.

All of this applies to audio formats as well - WAV, FLAC, and WavPack are all lossless (though the latter can support lossy as well), while MP3, AAC, Vorbis, and Opus are all lossy.

And it applies to video formats as well, but lossless video tends to be really large (luckily this was a short video clip).  ffv1 and huffyuv are two examples of lossless video formats (I used the latter during my conversion process) while formats like xvid, h.264, h.265, vp9, and av1 are examples of lossy video formats (though most of them have lossless modes as well, but the lossless modes have much worse support).


Since my initial process with the frames involved exporting the source GIF image to individual PNG frames at which point I batch-selected and deleted the repeating frames, no quality was lost because PNG is a lossless format.


Now technically converting color space between RGB (used by GIF and PNG) and YUV (used by 99% of all video formats as well as JPEG) is a lossy process, but the difference is pretty negligible, and my end result was going to end up being YUV anyway so I had no qualms about the RGB->YUV conversion process occurring. Admittedly though, doing this process multiple times can cause a build-up of rounding errors, but this is pretty easily mitigated by just editing at a higher bitdepth (e.g. 10bit) than what you're exporting at (e.g. 8bit) - a process very common in audio editing (e.g. editing at 32bit float or even 64bit float, and then exporting at 24bit or 16bit).

Similarly some video formats like the huffyuv also use chroma subsampling (reduces the resolution of color information) which, similar to GIF's 256 color limit, can be a lossy process if the source isn't already using chroma subsampling (source was 4:4:4), but again the end result was also going to be chroma subsampled to 4:2:0 anyway (99% of all digital video formats use chroma subsampling whether 4:2:2 or 4:2:0; JPEG by default also uses 4:2:0 but also supports 4:2:2 and 4:4:4).

[3deroticer]

good information here! If I make a batch PNG render and then take out the duplicate frame I still have to deal with renumbering all of them in order to use it for sequence animation. Has to be a better way!

[Rin Tohsaka]

Apologies for deleting and re-posting with additional content, but I forgot an important point regarding the removal of repeated frames - the frame rate!

Basically, if you remove regularly repeating frames, then you need to also make sure that the final frame rate is accordingly adjusted, otherwise your video or animation may play too quickly or too slowly.

If you remove every 5th frame on a 30fps video or animation, the result needs to then be set to 24fps.  Similarly, if there's only a repeating frame every 25 frames and the source is 25fps, the result would also then be 24fps.  Same goes for if the source is 30fps and there's a repeating frame every 4 frames, then the result would be 25fps.


There's a relatively simple math equation to figure out what the end resulting frame rate needs to be.

([source frame rate] / [frequency of repeating frame]) x ([frequency of repeating frame] - 1) = desired frame rate

So if the source is 30 and every 5th frame is a repeat of the frame just before it, then the equation looks like this:

(30/5) x (5-1) = 24

6 x 4 = 24

If I make a batch PNG render and then take out the duplicate frame I still have to deal with renumbering all of them in order to use it for sequence animation.

XnView has a built-in batch file renamer that not only works for images but actually works for anything (despite the program only being intended for images).

It's what I used to correct the frame numbers after removing the duplicates, and it's also what I used to "invert" the frame numbers for the copy of the frames that play in reverse.


If you're good with command line stuff, then the Linux terminal seems to also have the ability to do this sort of batch file renaming: https://linuxhint.com/rename_files_linux/

[Megan]

Emma Watson Shows Up Naked Exposing Her Udders Udders In Public

[John]

Nice work. To bad she is a crazy feminist.

[Megan]

Emma Watson big naturals!

[Megan]

Emma Watson’s Tits Got Bigger Than Ever