news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

New Intelligence Report

Editor: Editorial Department

【New Wisdom Introduction】AI photo-generating tool Flux has already taken the Internet by storm. This couple photo is realistic and delicate, with perfect lighting, texture, and hair. Video, sound, lip shape, AI is evolving to be more and more perfect!

If you don’t understand, just ask: What things on the Internet nowadays are still true?

Flux shocked the Internet

Today, the open source text-based graph model Flux has already taken the entire Internet by storm.

The following group photos were all generated by AI. This blogger was so shocked that he doubted his life.

The front close-up with a large aperture shows no flaws in the facial lighting, muscle texture, and hair.

Note that even the background characters are very natural and there is nothing wrong with them.

What if the lens is farther away and the light is darker? That's also natural.

The contrast between the two characters in light and dark creates the light and shadow texture of world famous paintings.

Just ask whether it is detailed and realistic, right?

Flux is not afraid even if there are more people.

Whether there are three, four, five or even more people, the picture is still impeccable.

The picky netizens are still trying hard to find tiny bugs.

The easiest thing to tell at a glance that it's AI is undoubtedly the text on the logo.

I felt the AI's efforts to get away with it.

There are also some details, such as the AI ​​does not understand what human hats and necklaces are used for, so there will still be loopholes in the drawing.

By the way, a year and a half ago, the most popular couple photo on the Internet during Midjourney V5 looked like this:

A large wave of secondary creation is coming

Now, the whole network has been swept by the Flux raw image storm, and people unanimously sigh: Flux has brought AI raw images to a new level.

Every time we think AI-generated images can’t possibly get better, they prove us wrong.

In the words of this blogger, AI is getting out of control and Pandora's box is being opened!

In particular, Flux's superb image creation capabilities and open source playability also provide a huge space for the creation of various secondary creations, videos, and voices.

There is no need to talk about these TEDx speakers that have gone viral.

This netizen used Flux, KeLing AI and synclabs to make a video of a YouTube celebrity blogger.

Although there are still traces of AI, the progress in images and videos is already amazing.

The author stated that his goal is not just to create an influencer, but to produce automated advertisements, YouTube, TikTok videos, instructional videos, marketing, lectures, and more.

Even when AI gets fast enough, it could be generated in real time, making it possible to FaceTime AI friends or AI therapists.

Yes, if there is one thing that is most terrifying about AI, it is its speed.

It only takes a few seconds or minutes to render an AI short film. There is no doubt that AI is going global, and everyone is optimistic about it.

This netizen said frankly: After introducing Flux.1 and Midjourney into AI videos, although they are not perfect, they are already the best AI works he has seen so far.

This blogger combined Flux and LoRA and found that LoRA also has a good processing effect on realistic images and painting/art images.

Flux.1 and LoRA also work well for animation generation and can be run on a single 4090.

This netizen said that he made two perfume ads in less than an hour. He said frankly: We are close to the singularity of AI video.

AI super evolution, in just 18 months

After watching a recent AI-generated video, an associate professor of AI at the Wharton School predicted that AI will complete its evolution in 18 months.

The reason for this view is that the speed at which AI models evolve is beyond imagination.

For example, the following pictures of otters using wifi on an airplane show visible improvement within a week or two.

Let’s take a look at what AI has evolved into after more than a year.

Not only is the image of the otter more realistic and adorable, but the hand movements when operating the phone are also flawless.

Let’s compare the speed of evolution of the same product: there is a very obvious improvement between MidJourney v3 and v4.

Musk's face was changed in one second, and the lip syncing was not exposed

Not only that, a recent GitHub study that went viral claimed that you can change your face during live broadcast with just one photo.

In the video below, Musk himself is seen putting on glasses and starting a real-time, delay-free live broadcast.

It’s to the extent that even Musk’s mother could be fooled.

There is also the big guy LeCun, who was also used by netizens to change his face for live broadcast.

Currently, the project has received 14k stars on GitHub and is on the Trending list.

Project address: https://github.com/hacksider/Deep-Live-Cam

At the same time, various lip-syncing technologies, such as ReSyncer, also make the mouth shapes of AI video characters extremely natural.

At this point, AI has made the entire workflow work! From now on, no matter how realistic the images we see on the Internet are, we may have to question them.

Increase saturation, AI raw images look timid

So, is there any way to identify traces of AI with a keen eye?

Taking advantage of the recent wave of AI-generated images, Deedy, one of the former founders of Google Search, proposed:

The best way to identify AI images is to increase the image saturation and carefully look at the microphone interface and teeth.

For example, if you set the saturation of a photo of a recently popular TED speaker to 200%, you can see his terrifying teeth.

It is worth mentioning that the code of the recognition tools was written by Claude and is publicly available.

Portal: https://claude.site/artifacts/6890e3d7-e65e-41ff-a7d4-3ccb38040b46

However, when we tested it with another AI-generated picture of a TED speaker, there were no flaws.

In this image, the teeth are not weird and the color is more consistent in the main color area.

Deedy said that if JPG compression is applied to real images, this consistency may be destroyed.

Netizen: I can only unplug the network cable

When we are in a "Truman Show" surrounded by AI, where is humanity's last resort?

Maybe it's time to unplug the network cable.

References:

https://x.com/AngryTomtweets/status/1822203767728591350

https://x.com/deedydas/status/1822665923775611374

https://github.com/hacksider/Deep-Live-Cam?continueFlag=4be7aad2ca0a560d6f9019228a8b2d3e

https://x.com/emollick/status/1822774265390985401

https://www.reddit.com/r/singularity/comments/1eo4sne/single_image_to_live_stream_deep_fake_deeplivecam/