Bai Jiao Hengyu sent from Aofei Temple

Quantum Bit | Public Account QbitAI

In Musk’s “latest live broadcast”, he didn’t button up his polo shirt properly, but was waving around with a cellphone shining on him.

Seeing his head turning left and right, with a smile on his lips, AK couldn't suppress it.

Why?!

The answer is revealed - this vivid Musk is not Musk himself at all!

It is generated by the latest AI face-changing project:

Just one photo, you can change your face and do live broadcast。

I have to say, Musk is pretty scary, a bit like the weird uncle in a horror movie~

Just install it locally to achieve real-time without delay, and it also supports real-time preview.

In other words, whether it is an online meeting or live streaming to sell goods, anyone’s face can be used.

As soon as you open the conference software, you will see 30 Musks having a video conference; as soon as you open the live broadcast room, you will see 100 Dong Yuhuis handing out large coupons.

This project is calledDeep-Live-Cam, without the need for dedicated hardware acceleration.

Although AI face-changing is nothing new, the project became popular overnight because “one photo can steal your face for live broadcast” and the effect is very good, and it has been open sourced.

Go straightGitHub Hot List No. 1, 24 hours of raging waves 1600+ stars.

Among the netizens watching, some were excited:

I'm laughing so hard that from now on I don't need to be present in person for any video conference hahahaha!

Some people also sighed that the era when even the elderly can make money through AI live streaming has finally arrived. This day came too fast like a tornado:

I thought this day would come, but I never thought it would come so soon!

Real-time face-changing live broadcast fraud

Have you seen the scene where Musk is in a rental house wearing a sweat shirt?

As soon as this project came out, everyone started to play with it like crazy, and currently it has accumulated more than 5,300 stars on GitHub.(Note that the introduction specifically states that it is uncensored)

For example, this netizen brought the repo and gave it a high evaluation.

It worked almost flawlessly and was completely live. It took me only 5 minutes to install.

The reason why the installation was so fast was that he later revealed his computing configuration, which was indeed top-notch.

He changed his face to look like a young Zuckerberg with an unshaven beard. Maybe this is what he will look like when he gets old.

Then he transformed himself into the handsome uncle George Clooney, but with a playful and smiling face.

And Hugh Grant.

I believe that careful friends have also noticed why these three have the same hairstyle and beard.

This should also be a small bug of this software: hairstyle cannot be changed easily and beard is difficult to hide.

Therefore, it cannot solve the needs of some bald people - there is no way to grow hair by replacing the face...

The author also noticed the risks behind this software and made a special statement at the beginning of the project.

First of all, this software aims to contribute to the artificial intelligence media industry to help bloggers and others complete tasks, such as animating custom characters, using characters as clothing models, etc.

To avoid unethical applications, it has a built-in check that prevents the program from processing inappropriate content.

The author also admitted that the project may be closed or watermarked in the future. He also suggested that if you use real faces, you should obtain the consent of the person concerned and clearly state that it is a DeepFake when posting content online.

The author then shared some tips on how to eat it. He first said that he（CPU）More likely to run, but it will be very slow.

Four simple steps: First, make some configurations: python (version 3.10 recommended); pipgit; ffmpe; visual studio 2022 runtimes (windows); clone the repository, then download two models: GFPGANv1.4 and inswapper_128_fp16.onnx, and put them in the models folder.

Finally, install the dependencies; if you want to use GPU acceleration, the author also provides further instructions.

You can use it after the installation is complete~ However, a model of about 300M will be downloaded when it is run for the first time.

As shown in the figure, select an image of a face and the target image or video, and then clickLive。

The predecessor of this project was calledroop-cam, take a video and replace the face in it with the face of your choice.

You only need to prepare a required face, such as an image of Black Widow. The whole process does not require a dataset or training.

Live mode is also supported.

Virtually indistinguishable photorealism

As the open source time increases, discussions among developers and netizens are also increasing.

Many people think this is "crazy" because a project with such a low operating threshold and such good results is actually open source.

Another group of people who think this is "crazy" are worried and feel that it has become even more difficult to distinguish between truth and falsehood in this world.

The problem that the latter is worried about is obviously facing each of us.

AI face-changing - whether it is live broadcast, video conferencing, pictures or voice, everyone who surfs the Internet now needs to open their eyes and carefully distinguish it.

Take the latest top-tier AI painting trend FLUX for example!

This is the first project officially announced by the original team of Stable Diffusion. Even AI master Kapasi enthusiastically reposted it to express his appreciation for its results.

Whether it is hands or text, the effect is very realistic and shocking.

So some Twitter users used AI generation tools such as FLUX and Keling to create a 100% AI YouTube anchor.

Netizens were moved to tears after watching it. "Why is AI so realistic? Don't we know what real people look like?"

No wonder some people sigh, is this the legendary “Silicon Valley meets the Uncanny Valley”?

The Internet of the future will definitely be a super weird place… although it’s already weird nowadays aaaaaaaaaaaaaaaaaaa.

In this regard, Quantum Bit can only say——

If, add thisLip Sync Model ReSyncerWoolen cloth?(Yes, there is a dog head)

ReSyncer released a comparison between himself and Heygen in lip syncing effects: Taylor Swift once again demonstrated her Chinese speaking skills.

What is the perception?

I think everyone resonates with this Twitter user to some extent:

AI technology has ushered in the era of truth and post-truth.

Speaking of which, I have to put some recent news:

3 days ago(August 7)The Xialu Branch of the Huangshi Public Security Bureau in Hubei Province announced that it had cracked a case in which AI face-changing technology was used to unfreeze social accounts and resell them for profit.

The three men used AI to swap faces in some videos, and then used the swapped dynamic faces to unfreeze some accounts that had been frozen by the platform.

In the end, the three of them used AI face-changing to unlock more than 2,000 social accounts and made illegal profits of more than 400,000 yuan.

Who says that the fraud risk brought by AI only comes from directly replacing faces?

However, it is an inevitable trend for AI generation and AI face-changing to move towards almost indistinguishable photo-level realism.

what to do?!

Some people are already getting impatient and have stated in the discussion forum that they are developing a deep fake detection plug-in for the codec used during rendering.

Let's cheer for him...

Reference Links:
[1]https://github.com/trending
[2]https://www.reddit.com/r/singularity/comments/1eo4sne/single_image_to_live_stream_deep_fake_deeplivecam/
[3]https://x.com/levelsio/status/1822067118914879808