Bowen is from The Temple of Convi
Qubits reports | Official account QbitAI
What if you have a disabled hand but still want to have your own two-dimensional waifu?
No problem.
True, just have a hand:
I have to say that this AI is a bit of a thing, and it is indeed painted very much like my future wife!
Yes, as long as you can give a sketch, AI can turn it into a two-dimensional beautiful girl, and say: more grass can do.
This AI, developed by the official technician "Amateur" of the Japanese illustration website Pixiv, once topped the Japanese trend list, and now the number of likes has approached 160,000:
After this, the author also added 512 parameter sliders, which can modify the painting style, face projection, shape and other directions, and the detailed settings have even reached 8192 types:
Netizens in the comment area called out to this:
Stop! Don't let artists become obsolete anymore!
Handicapped parties can also draw
The comments section below is now occupied by three camps.
Some of them happily choose a new wife from AI painting:
I thought that a statement such as "I was surprised to have my favorite face appeared in an instant" would fill the comment area, but it turned out not. So I posted a few likes out, so cute...
Ah, by the way, this kid was pushed by me.
There's also the divergence from this little AI, seriously thinking about whether this technology is robbing the illustrator's job.
It's like a personal illustrator at the front of the comments section with mixed feelings:
If someone asks me if this will cause me to lose my job as a painter, my answer is no. However, for someone like me who works hard, the technology does feel a little indescribable to me.
Some people also think of the previous AI who studied all the works of the manga god Osamu Tezuka, and then "created" a new work in the style of Osamu Tezuka, "ぱいどん".
For this trend of technology interfering in creation, these netizens said that they do not know whether this is lucky or unfortunate for real artistic creation...
But there are also many moderates who say: You are thinking too much.
Just like the world's first princess highness, the virtual kaehime Hatsune mirai, her appearance did not reduce the real singer, right?
On the contrary, many young people fell in love with music because of her.
So no matter how far technology develops, there will be no "no need for painters" to happen.
Some commentators also firmly believe that "technology is just a tool".
The work of being an original artist will not be affected, but the productivity of animation will be improved.
Of course, in addition to these slightly fierce debates, because the author has only released a demonstration video so far, a large number of netizens have also appeared in the comment area to brush the screen for open source and demo.
Development based on StyleGAN
Although the author did not reply directly, we have also introduced many sketch generation techniques and faces in previous issues.
And this AI should also be based on the further development of StyleGAN.
This algorithm that came out of nowhere in 2019 can be said to be a big step forward for face generation technology.
Although the previous GAN was good at the resolution and quality of the generated images, the researchers still could not understand the image generation process in more depth, such as how the random features of the image were controlled.
StyleGAN, on the other hand, redesigned the generator network structure of the GAN:
Starting with the learned constant input, the generator adjusts the image "style" of each convolutional layer based on the subtextile code, so that it directly controls the image features and thus the entire image generation process.
At the same time, because this process can be combined with the noise directly injected into the network, directly change the random attributes in the generated image, so you can also see that the whole work based on StyleGAN can be said to be endless.
The author is from Pixiv's official technical team
In fact, we also mentioned at the beginning that this AI is only an experimental amateur development of the author, and the developer himself is from Pixiv's official VRoid team.
The team's main product is Pixiv's 3D character-making software launched in 2018, VRoid Studio, which focuses on "creating 3D models in 2D".
As such, the artist can paint directly on the software-preset 3D model and adjust the parameters according to their own needs, and the results of the adjustment will be reflected on the 3D model in real time:
Perhaps it is related to the profession, the developer for their own amateur "sketch to comic" AI, also said that in the future will also add 3D model generation function, to achieve from sketch step to three-dimensional.
In this regard, I can only say: GKD.
Reference Links:
[1]https://twitter.com/t_takasaka/status/1477633104928178176
[2]https://twitter.com/t_takasaka/status/1478346974042923012