The AI Video Thread

[User 77373]

I've only got the 1 video so far but may as well start a thread for AI-generated or AI-converted videos. 

I've finally gotten around to playing with txt2vid. Here's where I'm at so far:
1. Generating a txt2img video


2. Use the video as a batch input for img2img with controlnet (openpose), keeping the seed constant.



3. Add interpolation frames and adjust the framerate using ffmpeg (not part of the webUI).



It's uncanny as hell but it's way better than I expected to be able to produce. It didn't take as long as I expected to run things either.

[User 77373]

IF dailymotion isn't working for you the last one is also on tumblr if that helps - (link)

[Cheshire]

Great work! It is weird but shows such promise.

How many images of the face do you have? If you can get a bunch (hundreds) of pictures of her face you might have some luck with deepfakes to smooth them out

This is the best single thread we have, although it is in the morphs section. Should probably be moved to AI:

https://preggophilia.com/pregnant-deepfa...r-thread-t

I’m not sure if deepfakes as they are can fix the belly button issue. But maybe someone on the forum knows a bunch of linear algebra.

[Cheshire]

Now that I give this more than a moments thought… you may need to paste in a video of real human’s face and then replace it with your model. Which actually seems like a bunch more work, not a bunch less.

[Fyra yu Chan 1999]

I've only got the 1 video so far but may as well start a thread for AI-generated or AI-converted videos. 

I've finally gotten around to playing with txt2vid. Here's where I'm at so far:
1. Generating a txt2img video


2. Use the video as a batch input for img2img with controlnet (openpose), keeping the seed constant.



3. Add interpolation frames and adjust the framerate using ffmpeg (not part of the webUI).



It's uncanny as hell but it's way better than I expected to be able to produce. It didn't take as long as I expected to run things either.

Wow amazing

[Azmarith]

Can you create a tutorial?

[User 77373]

I'd love to - for that and for how I do pics as well - but I have not much spare time so don't hold your breath.

the quick version for stage 1 and 2 is to get the automatic1111 webUI running, install the txt2vid extension, make videos until you get one that doesn't totally suck, then use the batch option in the img2img tab with the folder of the output frames as the input directory, enable controlnet, and use a non-random seed.

For stage 3, good luck. I can barely get ffmpeg to work at all, let alone instruct others.

[FranticScribbler]

We need more pregnant waddling videos. It's surprising to see ai video as a thing now