AI videos getting closer

[Nimos]

So Lumalabs released their video generator tool called Dream Machine so people can try it.

It is pretty impressive I think. I took my image here from the forum and told it to create a video of a hamster eating a flower.

This is the original image I uploaded.

And from that and the prompt it created this video:

There are some flaws in it, but what is extremely impressive is that it can actually generate the whole hamster as it moves away from just this single image and the consistency is also pretty good I think.

This is just based on a prompt: "illuminance Alien bugs swimming underwater, beautiful color floral"

It is a bit slow at the moment because there are so many people trying it, but if you want to give it a go it is free at the moment and allows you to create a few videos.

Lumalabs Dream Machine

 

[Eddi]

I dislike generative AI but these are good for what they are

 

[Nimos]

I dislike generative AI but these are good for what they are

You can like or dislike it, I can see arguments for both, but it is impressive I think as a technology.

One strong argument for it is that it allows more people to create their vision of something. Whether you want to make a science fiction movie, or a movie about weird alien creatures, music or just images for whatever project you might have, that would otherwise be impossible. Either due to money or simply time.

I think a lot of people have good ideas but simply do not have the means to realize them or simply not knowing the right people to make them come true, AI as a tool allows more people to do things, and I like that.

 

[Nimos]

Was waiting for another video of a hamster, but this one I forgot to use the image, but this is pretty damn good I think:

 

[McBell]

 

[McBell]

I present another video explaining how AI video is already here:

I do not present transscripts because you really do have to see it in order to understand what is being talked about.

Edit:
Ok, I changed my mind about the transcript

Conclusion​

So deepfakes are here, and they're seemingly here to stay.​

We are entering a new chapter of how humans consume and relate to information.​

It's a chapter as significant in my mind as some of the other big chapters,​

like when we learned to capture light and record it.​

It changed the world!​

Photographs could be used in court as proof.​

They could be used in science to capture and understand the world around us.​

In journalism, it allowed us to show, not just tell.​

And when we made those images move, we could capture more truth, more proof, more evidence.​

And then we could connect, and spread those images far and wide.​

Truth and evidence coursed through the world's internet.​

This evolution of the moving image and its uses in the connected world is taking an unexpected turn,​

away from being the bedrock of evidence, and towards the foggy territory of deception and confusion.​

We're gonna have to navigate this chapter carefully.​

We're gonna have to push even harder to know where our images came from, who made them, and for what purpose.​

And we're gonna have to resist the urge to believe everything we see, no matter how real it looks​

 

[Nimos]

I do not present transscripts because you really do have to see it in order to understand what is being talked about.

There is and will be even more of this in the next couple of years. I think the big difference between the one I used and those you linked, is that they are more difficult to make.

Mine is simply made through a text prompt and/or an image and everyone can do it, it is easier to do than writing a post here on this forum and what you see now is obviously as bad as it is ever going to be.

I think we can expect the same progress with videos as we have seen with images and music, which can already make things where you can't tell the difference.

So surely there are some potentially very concerning uses of this technology when it comes to spreading misinformation. Especially when it comes to political stuff I would assume this could be a big problem.

 

[McBell]

There is and will be even more of this in the next couple of years. I think the big difference between the one I used and those you linked, is that they are more difficult to make.

Mine is simply made through a text prompt and/or an image and everyone can do it, it is easier to do than writing a post here on this forum and what you see now is obviously as bad as it is ever going to be.

I think we can expect the same progress with videos as we have seen with images and music, which can already make things where you can't tell the difference.

So surely there are some potentially very concerning uses of this technology when it comes to spreading misinformation. Especially when it comes to political stuff I would assume this could be a big problem.

The videos I presented are over a year old.
Your video is taking a still photo and animating it.
With low level programing doing the work.

The second of the videos I presented show what could be done a year ago with medium to almost high level software.

Imagine what can be done now with high end software and hardware?

 

[Nimos]

The videos I presented are over a year old.
Your video is taking a still photo and animating it.
With low level programing doing the work.

The second of the videos I presented show what could be done a year ago with medium to almost high level software.

Imagine what can be done now with high end software and hardware?

Only the first video of my avatar is based on an image.

The alien one and the last hamster are not, they are fully generated from scratch based on the prompt I gave it.
It is trained on videos of whatever and based on these the model tries to generate what it is told.

Whereas those you linked require a lot more effort from the creator as far as I know.

 

[McBell]

Only the first video of my avatar is based on an image.

The alien one and the last hamster are not, they are fully generated from scratch based on the prompt I gave it.
It is trained on videos of whatever and based on these the model tries to generate what it is told.

Whereas those you linked require a lot more effort from the creator as far as I know.

The second one I posted goes into a "geek mode" explaining how they use a creator AI and a Detective AI to greatly improve the quality of the fakes.

 

[Nimos]

The second one I posted goes into a "geek mode" explaining how they use a creator AI and a Detective AI to greatly improve the quality of the fakes.

It is crazy and will become a much bigger issue as time goes by.

But still, I think the one they are using in the video is more trained on a specific task such as replacing the face rather than being generative because he doesn't seem to have any of the usual issues, which you can see in the example below.

This was supposed to be a: "Wolf cub playing in the snow" and it is horrible.

Also in the videos, he doesn't have issues with his hands, which is the most clear giveaway at the moment. Also, the skin is very smooth in the video which is also a giveaway, but it does look very good, no doubt about it, and I don't think people would notice if they weren't paying attention to it. If I just saw a video with a deep fake about whatever topic, there would be two primary things that would alert me, only with famous people that I knew.

1. The person saying something that is out of character.
2. If the quality of the video is bad.

But for regular people or those I'm not that familiar with, I think I would be fooled most of the time.

Again, what they are doing in the video is not that easy for a normal person, whereas again the one I use is. And it is just a matter of time before it gets better.
The next issue is obviously that you can also fake text, images and audio extremely easy now, which only makes it even harder to detect and also to spread misinformation.