The recent trend of Drawing AI is tremendous, and it has evolved every day. These AI focuses on how to generate images by receiving instructions (called prompts) in words from users, but can be used as imaging, which receives the original image and finishes it.
- As expected, TOMB RAIDER I is a low-poly
- The key is to upload people
- Anniversary version that can be said to be smooth when compared
- The conclusion is satisfactory result
Let ## AI make the low-poly character high definition!
So, if you pass the screenshot of the old work to AI and finish it in a real or high-resolution art style, the illusion that was complemented in the brain at that time looking at the character of the low polygon. Isn’t it? So, this time, I will give AI to AI from the Tomb Raider series and try to complement the image.
There are two targets: Tomb Raider I , which is sold at Steam, and Tomb Raider: Anniversary , which is slightly higher graphic in 2007. The expectation before it started was that the anniversary version was so faint that the feeling of jerks was diminished, so it was anxious whether it would be effective as expected.
As expected, TOMB RAIDER I is a low-poly
The drawing AI used this time is Stable Diffusion. As an environment, Stable Diffusion Web UI was used, and the model was selected for the original SD-V1-4.CAPT which is not particularly tuned. The output result changes considerably depending on the model used, the prompt to the AI, and the detailed settings, but let’s try it without a prompt, especially the settings.
First, from Tomb Raider I, the camera moves automatically according to the character’s movement, so I had a hard time turning the front, but one of them could be used.
If you think of this as the technical skills of the time, I feel the progress of the times. By the way, aside from the snow in the background, I would like you to recognize the tank top and both guns. What is the result?
Something like a contemporary art style with a hospital or something motif has been created. I tried it several times under the same conditions, but it turned out that there were many patterns such as objects that are not well understood in the white gray background, and that the images are not recognized as expected.
As expected, it is too devastating, so adjust the parameters that determine how much the original composition is maintained, and write down keywords such as LARA CROFT and TOMB RAIDER at the prompt. 。
This is a relatively good one I tried several times. The direction is definitely the same, but no matter how many times it is generated, the face will collapse, and the character is too distant in the first place? As the question has arisen, I will try the original image and try it.
The key is to upload people
I’m worried because it’s quite low resolution, but what happens?
I got an impactful result. Isn’t this a success? When this one was generated, the prompted keywords such as Photo and Realistic may be affected. You can also see the movie Angelina Jolie starring the movie version.
If you adjust the parameter to the original image, like this.
I’m a little scared of my eyes, but it can be a piece that tastes like 3D animation.
Anniversary version that can be said to be smooth when compared
Similarly, let’s generate the anniversary version while narrowing down the adjustment. Based on the previous results, it seems that good results can be obtained.
Click here for the result of putting the start screen as it is.
I would like to evaluate the attitude of AI, which is trying to hold menus and titles, but if the person is far away, the face will easily collapse. It seems that it is necessary to trim and upload it.
Click here for the result of trimming and trying.
I got a pretty wonderful 1. Looking at the resulting results, it may be the image of AI’s LARA CROFT that eyelashes seem to have a strong eye power.
The following is one piece in the game with sunglasses. There was also a problem that if you do not put GLASSES in the prompt, you will not recognize sunglasses and will be treated as something that is not.
The conclusion is satisfactory result
This time, it was often the result of a photorial generation, and one realistic piece was obtained compared to the latest game screen. Of course, that graphic does not move because it is one image, but it can be said that the drawing AI has been well confirmed. Nevertheless, there is a problem that it is not versatile yet, and it is necessary to make a prompt to get this intention.
As a writer, Lara Kraft’s image is a woman who combines strongest and beauty, but even if you look at a large amount of generations, it tends to be output with a dignified expression, and that low polygon is convinced. I am very satisfied that it is converted to a beautiful graphic in the form. The anniversary version was also more effective than expected.
Of course, the results of this time are only those that AI learned and generated, and just selected and introduced. Please note that each individual does not deny the illusion he saw on the Low Polygon character that day.
(Note) I want you to make the character of that game and this game a high resolution! If you have any, please feel free to in the comment section. We look forward to your opinion.