2024/08/10 12:40
The live version of the deepfake is too scary! Real-time video face swap for a single photo, 5 minutes to install without picking hardware
2024/08/10 12:40
Musk's "latest live broadcast", he didn't button his polo shirt well, and dangled it around with a mobile phone.
Seeing his head turn left and right, and the smile on the corner of his mouth, AK couldn't suppress it.
Why?!
The answer is revealed - this vivid Musk is not Musk at all!
Rather, it is generated by a new AI face-swapping project:
With just one photo, you can change your face for a live broadcast.
I have to say that such a Musk is quite scary, a bit of a ghost movie and a strange taste~
Install it locally, you can achieve real-time without delay, and it also supports real-time preview.
In other words, whether you are holding an online meeting or live streaming, you can use anyone's face.
As soon as I opened the conference software, 30 Musks were having a video conference; As soon as the live broadcast room was opened, 100 Dong Yuhui were handing out large coupons.
This project, called Deep-Live-Cam, does not require dedicated hardware acceleration.
Although AI face swapping is not a new thing, because "1 photo can steal your face to live broadcast", the effect is good, and it is also open source - the project became popular immediately.
Go straight to the first place on GitHub's hot list, with 1600+ stars in 24 hours.
Among the onlookers, there are some who are excited:
I can't laugh anymore, I will participate in any video conference in the future, and I will never have to attend in person again hahahahaha!
Some people also sighed again and again that the era when even the elderly can use AI live broadcast to make money has finally come, and this day is coming too fast like a tornado:
I thought it would come one day, but I didn't expect it to come so quickly!
Have you ever seen Musk wearing a sweatshirt in a rental house?
△ "Musk" stared at the screen seriously
As soon as this project came out, everyone started to play like crazy, and so far GitHub has accumulated 5300+ stars. (Note that the profile is also deliberately stated to be unreviewed).
For example, this netizen brought a repo and spoke highly of it.
It works almost flawlessly, and it's completely real-time. It only took me 5 minutes to complete the installation.
The reason why the installation is so fast, in the back he also revealed his calculation configuration, which is indeed very top.
His bearded little Zazakberg may have become like this when he gets old.
Then he transformed into a handsome uncle George Clooney, but a hippie smiley version.
And Hugh Grant.
I believe that careful friends can also see why these three have the same hairstyle and beard.
This is probably a small bug in the app: the hairstyle can't be easily changed, and the beard is hard to hide.
As a result, it doesn't solve some of the baldness needs – there's no way to grow hair by changing your face......
It is also aware of the risks behind this software, and the author hereby declares it at the beginning of the project.
First of all, this software aims to contribute to the AI media industry, helping bloggers to complete tasks such as animating custom characters and using them as clothing models.
To avoid unethical applications, it has a built-in check feature that prevents the program from handling inappropriate content.
The author also admits that in the future, the project may also be closed or watermarked on the output. It is also recommended that if you are using a real face, it is recommended to obtain the consent of the person concerned and clearly state that it is a deepfake when posting content online.
The author then shared tips for eating. He started by saying that he was more likely to run on his local computer (CPU), but it would be very slow.
Four simple steps: first do some configuration: python (recommended for version 3.10); 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 dependencies; If you want to use GPU acceleration, the author also provides further instructions.
After the installation is completed, you can use it~ But when you run it for the first time, you will download a model of about 300M.
Select an image of a face as well as the target image or video as shown in the image, and then tap Live.
The predecessor of this project was called Roop-Cam, and it was a video that was shot and replaced with a face of your choice.
You just need to prepare a face you want, such as an image of the widow. No dataset is required for the whole process, no training is required.
Live mode is also supported.
As the open source has been open for a long time, the discussion between developers and onlookers has also increased.
Many people think that this is "crazy", because the project with such a low operation threshold and such a good effect is actually open source.
The other group of people who think this is "crazy" is worried, and thinks that the world is more difficult to distinguish between real and fake.
The latter concerns are clearly in front of each and every one of us.
AI face swapping - whether it is live broadcast, video conference, pictures, or voice, everyone who is surfing the Internet nowadays needs to be carefully identified with wide eyes.
Let's take the recent AI painting industry, the latest top FLUX!
This is the first project officially announced by the original team of Stable Diffusion, and even the AI god Kapasi passionately forwarded it to express his appreciation for its effect.
Whether it's a hand or a text, the effect is very real and shocking.
△"Photos" generated by FLUX
So there are Twitter friends, who used AI generation tools such as FLUX and Kelin to make a 100% AI YouTube anchor.
Netizens cried after watching it, "Why is AI generated so realistically, don't we know what real people look like." ”
No wonder some people sigh, is this the legendary "Silicon Valley meets the uncanny valley"?
In the future, the Internet must be a super strange place...... Although it's already weird now, ahh
In this regard, qubits can only say -
What if, with this lip sync model, ReSyncer? (Yes, this has a dog head).
ReSyncer released a comparison between himself and Heygen on the lip sync effect: Mildew reproduces the magic skill of speaking Chinese.
What is the look and feel?
probably more or less resonates with this tweeter:
AI technology has brought the era of truth and post-truth ·.
Speaking of which, we have to put a recent news:
3 days ago (August 7), the Xialu Branch of the Public Security Bureau of Huangshi City, Hubei Province, announced that it had cracked a case of using AI face-swapping technology to unfreeze social accounts and resell them for profit.
The three men performed AI face swapping on some videos, and then used the dynamic faces after face swapping to unfreeze some accounts that had been frozen by the platform.
In the end, they used AI to swap faces, unlocked more than 2,000 social accounts, and made illegal profits of more than 400,000 yuan.
Who said that the risk of fraud brought by AI only comes from direct face swapping?
However, it has become an inevitable trend for AI generation and AI face swapping to almost indistinguishable photorealistic realism.
What to do?!
Some people can't sit still anymore and say in the discussion forum that a deepfake detection plug-in is being developed for the codec at render time.
This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com