Professor
Professional
- Messages
- 322
- Reaction score
- 124
- Points
- 43
This is not a working, untested and rather problematic idea, purely for informational purposes, for fun and discussion.
The idea came to me at 4:20 local time. Text to image neural networks were released one after another, pornpen came out, which at that time could only draw ugly Kronenbergs with a thousand boobs and three arms. And then I thought, what if...
I take photos of girls in bathing suits, "undress" them with a neural network, send them their own nudes and demand funds for non-distribution of the content?
You will say ugh and you will be right.
At that time, I started a white project, and if it had not been for it, I would hardly have done this, too hemorrhoidal and risky. But I remembered my school years and fake nude applications, still for Symbian s60 (I wonder if their distributors made at least a little money?). Maybe at least now it is possible to fulfill the dream of a sperm-toxic teenager?
An analysis of undressing services, unfortunately, showed that it is not that easy.
Existing services undress sloppily and with a lot of artifacts.
It is assumed that you can run Disco Diffusion on your machine, train it on nudity and undress with it. But this is already a task that requires model training skills. Ideally, it should not just undress the girl, but also change the background and lighting, so that it is not obvious that the photos are taken from Instagram and "photoshopped".
In general, I played with photos of girlfriends in one paid public service, admired the unsatisfactory results of their undressing. So bad that even pranking no one would not have worked, I gave up and went to undress a meat 3D woman, in any case more pleasant and useful for the mind and body.
As for the idea for the scheme, I see many problems that greatly limit its performance:
- complex preparation, you need to manually select photos, run them through the neural network, edit the results, send them to the victim, who will most likely answer: "I don't care, fuck off";
- a very limited audience, due to the complexity of spamming Instagram and the need for photos from certain angles and a bunch of other nuances;
- low efficiency, especially if the result will at least slightly give itself away as an edited photo;
- complexity and resource intensity. There are no public services with acceptable quality, and training the model yourself, running it on VERY powerful hardware - with such resources and skills, you can come up with something more profitable.
Perhaps it makes sense to roll out something like pornpen, only image-to-image, specifically for undressing, and with the functionality described above. I don't know how much such a project will cost, but there will be demand for it, there are enough wankers with money.
The idea came to me at 4:20 local time. Text to image neural networks were released one after another, pornpen came out, which at that time could only draw ugly Kronenbergs with a thousand boobs and three arms. And then I thought, what if...
I take photos of girls in bathing suits, "undress" them with a neural network, send them their own nudes and demand funds for non-distribution of the content?
You will say ugh and you will be right.
At that time, I started a white project, and if it had not been for it, I would hardly have done this, too hemorrhoidal and risky. But I remembered my school years and fake nude applications, still for Symbian s60 (I wonder if their distributors made at least a little money?). Maybe at least now it is possible to fulfill the dream of a sperm-toxic teenager?
An analysis of undressing services, unfortunately, showed that it is not that easy.
Existing services undress sloppily and with a lot of artifacts.
It is assumed that you can run Disco Diffusion on your machine, train it on nudity and undress with it. But this is already a task that requires model training skills. Ideally, it should not just undress the girl, but also change the background and lighting, so that it is not obvious that the photos are taken from Instagram and "photoshopped".
In general, I played with photos of girlfriends in one paid public service, admired the unsatisfactory results of their undressing. So bad that even pranking no one would not have worked, I gave up and went to undress a meat 3D woman, in any case more pleasant and useful for the mind and body.
As for the idea for the scheme, I see many problems that greatly limit its performance:
- complex preparation, you need to manually select photos, run them through the neural network, edit the results, send them to the victim, who will most likely answer: "I don't care, fuck off";
- a very limited audience, due to the complexity of spamming Instagram and the need for photos from certain angles and a bunch of other nuances;
- low efficiency, especially if the result will at least slightly give itself away as an edited photo;
- complexity and resource intensity. There are no public services with acceptable quality, and training the model yourself, running it on VERY powerful hardware - with such resources and skills, you can come up with something more profitable.
Perhaps it makes sense to roll out something like pornpen, only image-to-image, specifically for undressing, and with the functionality described above. I don't know how much such a project will cost, but there will be demand for it, there are enough wankers with money.
