The image generation AI "Stable Diffusion" has a function called "img2img" that generates arbitrary images using illustrations drawn for instructions and text for instructions.
For example, if the above illustration is given the instruction "A Red Classic Car" and an image is generated, the following results are obtained.
I applied this to a game screenshot and processed it.
Below is a screenshot of my Mazda MX-5. In addition to this, I will generate an image with the instruction "Mazda MX-5 Miata" to the AI.
The following image is then generated.
If the same wording is used again for the above image, the instructions would be as follows.
These results are very interesting, and we can say that they have the potential to make game screenshots more realistic.
If you are interested, please give it a try.
Hi 1918 Can You Make A Garuda Indonesia 747-200?? The Indonesians Would Be Very Happy!
Can you put the last image into the AI another time?
@among1 bot
I like somehow the suppoesed to be the Wright Airport has become a highway on grassland.
And thus, it reminds me to Forza Horizon
Cool
That's really interesting. The background the AI generates is quite amazing, and it would take little effort to photoshop the build onto the AI generated, or I guess "improved" background.
@1918 how to use it?
@1918 how i access the img2img function
This is cursed and cool at the same time
thanks
@1918
https://github.com/cedro3/others2/blob/main/Stable_Diffusion2.ipynb
@JeskoGoesVROOM
give link pls
nerd car
Yoooo that's pretty cool. Time to go spend like an hour messing around with it.