September 10, 2024
In a mobile app called Snow, a subsidiary company of Naver in South Korea, a backward country in AI, a user who tried to use a service that uses artificial intelligence to expand and draw the outside of a photo, such as a photo taken with his smartphone camera or a photo uploaded to the app by the user through his smartphone, was reportedly drawn in an obscene manner in the photo of a user who tried to use the service (out-painting), but was ignored and the photo he did not want was reported as obscene and deleted.
 |
| The Soda app of the issue is currently on Google Play |
 |
A photo of the Soda App outpainting results that was removed because it was mistaken for a user's own photo and pornography.
In addition, an app called Snow, created by the same Naver subsidiary, Snow, also used Faceswap AI feature that allows users to upload or take a photo of their face and save it, and then synthesize different colors and hairstyles on the photo also showed a nude photo of the user and the photo that was allegedly pornographic.
 |
| A naked photo received while using the Snow application (app) AI Hair Shop feature. The photo above is a mosaic of the original, Photo courtesy of a reader |
The reason why these problems occurred one after another is that the photos taken with the camera or uploaded by the user to the app are transformed by such functions using an artificial intelligence application installed on the server, but the problem is that those scumbags of the company, a subsidiary of Naver, implemented out-painting and face swap functions in conjunction with applications such as Stable Diffusion, which is open source, to easily implement such out-painting and face swap functions. In that case, it's a scumbag company that's all about the money.
Image
generation apps such as Stable Diffusion are distributed for free and used by
many people around the world, and it is known that it is a very popular
application for the creation of deepfakes and related artificial pornography
that has recently flooded the Internet, and while it is easy to install and
maintain the app, it is difficult to implement a proper control not to mass
produce pornography.
In the world
except Korea, artificial intelligence image generation applications such as
Stable Diffusion are installed for free and used by many individual users, but
in Korea, which has a language barrier and a rice wall barrier and is becoming
the Galapagos of global technology, it seems that it is provided in the form of
a paid service mobile app for general
users by linking these applications.
However, the
problem is that this kind of accident occurred because the uncontrollable open
source app was linked with a commercially paid app to provide these AI services.
Even in the
case of NAVER's subsidiary Snow, it should have provided this feature only with
an image generation model with NSWF filter that does not generate such
pornography in the first place, or had the relevant stable diffusion management
personnel in the company strictly manage the app.
After examining
the contents of related broadcasts and news articles, I strongly suspect that
the out-painting function, which draws the outline of the photo with artificial
intelligence, and the hair style change function (a typical face swap function)
were not implemented on the servers of Naver's subsidiary Snow, but by
connecting the 3rd party online service company through the API.
Currently,
image-generating AI applications like stable diffusion type are centered on AI
models that draw pictures or drawings according to text prompt commands, but also there are limited other functions
such as changing the picture into a different shape or expanding the picture
when you enter a picture or image. The function that draws a photo or picture
when a prompt command is entered is called tex2img, and the function that
outputs a picture when a photo or picture is entered is called img2imng.
img2img is mainly used to generate adult materials such as deepfakes but there
are not many AI models in this img2img function, and most models are trained to
generate adult materials that are not prohibited by NSFW.
In addition,
if a NAVER subsidiary called Snow installed such an app on its own servers and
provided related functions, it is inferred that the personnel who operates the
relevant servers in the company accessed these servers from time to time and
drew pornographic images, and that the usage of such applications by the
operators was so frequent that the pornographic images were synthesized by
reflecting this usage pattern. This is because these applications also have a
tendency to produce pictures or drawings based on their usage.
At this
point, I think the police should seize and search the servers of these
companies. And if it is true that these applications provide related functions
online through camera apps, the photos of the user's face that the user enters
or uploads online are stored in a specific storage folder on the server
where the stable diffusion is installed as input data.
Additionally, the details
of the artificial intelligence model that generates only the faces of specific
celebrities through App's image-generating artificial intelligence function
related to AI app such as Stable Diffusion are as follows: after collecting photos
of celebrity faces that are flooded on the Internet, training an artificial
intelligence model with those face pictures, installing it in the app, and
entering a few prompts related to the celebrity, the Lora (Low Level
Adaptation) artificial intelligence model is trained to generate only photos
with the celebrity's face, and images are generated using it. Therefore, there
is currently no way to prevent such an artificial intelligence model from being
used in such an artificial intelligence app.
Previously,
it took 500 to 1000 photos to create such a Lora A.I. model, and it is was
possible to train such a model by running a computer with powerful hardware for
several hours, but now, a personal gaming PC can with about 10 photos of the celebrity's
face and a few tens of minutes of training depending on the computer's
performance can train the model and there are already A.I. apps that make this
very easy, and the information is being shared on the Internet
about how individual developers can further develop these training apps and
make the models on their own personal computers with less expensive HW. Of
course, there is a language barrier, and in Korea, which is a technological Galapagos,
few people are doing this, so most of the people who are mass-training these AI
models with Korean celebrities' photos and posting them on related websites are
Chinese commies.
Now that the Pandora's box of image-generating
A.I. apps has been opened, there is no way to stop it no matter what. So, by
using an online service that changes a photo you take with A.I., you are giving
your face biometric information to these garbage companies that only want to
make more money. Therefore, it is really, really stupid to use an online
service that changes a photo with A.I., unless it is an image-generating A.I.
app installed on your own personal computer, and even pay for it with your own
hard-earned money is really absurd.
Do you
really think the ignorant scumbag reporters of the propaganda yellow journalism media are
going to realize the dangers of this and report it properly? I don't think so.
It's
pathetic to see such an AI backward country, where trashy companies that don't
even know the basics of AI are offering these dangerous generative image AI
apps online as a paid service without proper internal control, without proper
management, and without any hesitation after such an accident. What future is there
for AI in such a backward scumbag country?
This scumbag country can prosecute middle schoolers charged for making deepfake porn pictures out of curiosity but never prosecute big IT service companies doing same thing.