It is amazing if the image generation AI ``Stable Diffusion'' is used to make the first generation Virtua Fighter's kakukaku polygon high-quality to a modern real 3DCG



Stable Diffusion of image generation AI is a type of model called 'latent diffusion model', which not only generates images according to given instructions, but also up-converts low-resolution images to high resolution and improves image quality. It is also good at compressing images while preserving them. Using such Stable Diffusion, the image of converting the original Virtua Fighter , known for its kakukaku polygon characters, into a modern and realistic 3DCG image has become a hot topic.

Begone, polygons: 1993's Virtua Fighter gets smoothed out by AI | Ars Technica
https://arstechnica.com/gaming/2022/10/begone-polygons-1993s-virtua-fighter-gets-smoothed-out-by-ai/

Virtua Fighter is a 3D fighting game developed by Sega that first appeared as an arcade game in 1993. It uses polygon drawing for graphics, and is the first fighting game that expresses characters and stages with 3DCG models. However, in 1993, in order to move the 3DCG character in real time, the number of polygons of the character had to be small, so the appearance is quite crispy, based on the Virtua Fighter nearly 30 years ago. It was a big feature of the game. You can see how Virtua Fighter actually looks like by watching the following movie.

Virtua Fighter (Arcade, 1993) Real Arcade Hardware Capture-YouTube


CG artist Colin Williamson has announced an image of Sara, who appears in this Virtua Fighter, rewritten into a modern and realistic 3DCG with Stable Diffusion. The left is Sarah of the original Virtua Fighter, and the right is Sarah made real by Stable Diffusion. The number of polygons has increased at once, and the hairline, facial expressions, and the materials of the clothes he is wearing have become more realistic, and the sunset in the background has a more natural atmosphere.



After fixing Jackie who is the user of Jeet Kune Do, it looks like this. The bulging muscles and tank top are realistic, and the face looks quite old.



By the way, the appearance of Jackie of the original Virtua Fighter 5 is as follows, a much more youthful image than the Stable Diffusion version.

Virtua Fighter5 Official Website - Jackie Bryant
https://www.virtuafighter.jp/vf5/cha_08.html



And when Akira, which can be said to be the face of Virtua Fighter, is corrected with Stable Diffusion, it looks like this. With a very realistic Japanese face, I have the impression that you are more like Nakayamakin than Akira, and rather than the famous line 'You're 10 years early!' There is atmosphere.



Kage (Kagemaru), who uses Hagakure style ninjutsu, could not understand the bowl metal and chain mail wrapped around Stable Diffusion's forehead, so in the modified version it looks like a masked fighter rather than a ninja.



Lau Cheng is a user of Chinese kempo. Virtua Fighter's Lau looks severe and steep, but for some reason Stable Diffusion's modified Lau's face is lowered, giving the impression of a Chinese restaurant owner who is insanely good. The clothes are also very beautiful, and the size does not match, giving a somewhat cosplay-like atmosphere.



Geoffrey and pie also look very realistic. Jeffrey's left arm and body are slightly uncomfortable, but his face is super realistic. Rather than a martial artist, Pai has the impression of ``a person who just poses like this''.




Canadian-born professional wrestler Wolf looks relatively realistic. However, it is supposed that there was no characteristic tattoo on the face.



Hidden character Dural. Dural's face, whose whole body has a metallic luster, has the same facial features as a human being, but Stable Diffusion makes it look like a mysterious cyborg. Still, you can see that the smoothness and texture of the body have improved considerably.



Mr. Williamson said that he used i mg2img of Stable Diffusion web UI for AUTOMATIC1111 to correct it. Of course, Stable Diffusion is not an algorithm that can be corrected as desired in one shot, so it was necessary to repeat trial and error while making fine adjustments. It seems that the most difficult thing was the reproduction of clothes.

Williamson told Ars Technica, an IT news site, ``If you find a good prompt, run about 50 patches and choose the most interesting one. We also give orders to AI.”

Mr. Williamson also modified the character of the gun shooting game ' Virtuacop ' released in 1995 with Stable Diffusion.


Related Posts:

in Software,   Game,   Art, Posted by log1i_yk