[LOCKSUIT]

for picture


for sound
(see on his youtube channel, it copies your voice even better than these movies above!)

[LOCKSUIT]

"It’s interesting to note how ray-tracing converges with deep learning. There’s no more need to trace all the rays, deep learning predicts rays."

A nice history of image AI evolution:
https://blog.inten.to/welcome-to-the-simulation-dd0d8cb6534d

[HS]

Nice, its boiling the world down to the true essentials needed for efficient prediction. Maybe we could learn from the examples we've already created. All the creative things people engage in should leave footprints of the internal processes which created them.

Art could be thought of as distilled reality which is plastered on to the panorama of our imagination. If data is a symbolic representation of information, while art is symbolized understanding, then it follows that we have a complementary value detector for understanding. We just need to count and categorize which distillates of reality are important enough to be included in different kinds of art. Then we’d end up with a recipe, “to understand x it is most efficient to reduce reality to the following five essences…“

The advantage of essentials is they don’t require much bandwidth. If the devil is in the details then god must be somewhere in the blurs. IMO The Lord of The Rings movies are better art in SD, so in this case low band width actually simplifies understanding because it allows efficiency. These reality essences are tools in our mental toolbox. Leverage points for optimizing the proximate environment, and enduring into distant environments.

Life tries to persist. Life forms and life lines tend to persist due to the maintenance of equilibria. General Intelligence could be equated to the complexity of the equilibriums which it is able to maintain. So, the things included in art could be tools for creating equilibria, the use of which would enable efficient processes and create intelligent actions.

[LOCKSUIT]

If anyone here tries it, will you try one for me? I got a good pic to try but too busy on my pre-AGI planz...

[yotamarker]

where is the walkthrough ?

[LOCKSUIT]

Ah, AHHHHHHHHH, I have made that beautiful girl alive!!

just change the extension to .mp4

[LOCKSUIT]

woah, there wasn't an ear in the image !
just change the extension to .mp4

[8pla.net]

This is really very interesting!

No complaints, LOCKSUIT, just saying...

AIdreams supports uploading mpg videos.

And there are other cool options.

Do you have any examples that won't

attract the attention of the secret service  ?

Cite Source: Two Minute Papers
             URL:  youtube.com/watch?v=mUfJOQKdtAk

[LOCKSUIT]

No more for now.

[8pla.net]

How did you do it?

[LOCKSUIT]

Two Minute Papers has another older video from a month ago or so that shows the mona lisa head doing it really good, that is farthest i seen info pop outta tha image if you get me. Interesting to see what else it can do though. And will read the paper.

I just go to the link he gave:
youtube.com/watch?v=mUfJOQKdtAk
and click all buttons to run it. I used a dummy gmail account to do it cus it has accessssss which is creepy. At the folder part, you need the folder the code leads you too! Lol! Put it in your My Drive on gmail, and the content/etc/etc actually points to it there at root. Secondly, in the code below, you'll see the 2 name files to open, change name to select different files in the folder. There seems to be 3 different generators below, top one seemed most best by all regards if i'm correct.

[LOCKSUIT]

The question is, is is this the same tech as the following? This one looks better.


from the above paper:
"rather than by warping. The talking heads created by our system
can, therefore, handle a large variety of poses that goes beyond the abilities of warping-based systems."

from this thread's paper:
"generator network that warps the source image according to dense motion and inpaints the image parts that are occluded in the source image."

[LOCKSUIT]

specify, 1:27 lol

[8pla.net]

May I suggest we keep this topic active?

From the Description on YouTube, here

is open source Python 3.5 source code:

Demo for paper "First Order Motion Model for Image Animation"

[LOCKSUIT]

That's what I used.

The older method seems superior and they say why too, I posted it above.
Btw my source image to drive was only 1 image, you probably can feed it a video.
My testing seems to show it doesn't like rotating heads, though a watermark was in the way . It was ghostly ruined and dissipated/blurred.

The human brain should be able to super-resolution objects, change their color/colorize them, increase frame rate or extend a frame or extend the movie or inject a cutscene between 2 parts ex. "my cat [who blah blah....] ate food", rotate them, etc.