Can indies use this AI texture upscaling to boost their game content?

whose machine can run that though? Especially if you upscaled the entire environment to the same fidelity as the weapon? I don't even have the internet to run the video at your suggested resolution.

i think an important part of game art is knowing how and when to compromise, how to compensate for limitations of the medium, how to prioritize efforts around that.

I mean, if you have the means to achieve full photorealism and make that run at suitable framerate, go for it. But if you don't -- and most don't -- you have to adjust course. If you try to shoot for a goal that is obviously impossible to reach, you'll come up short. And it's gonna show.

Personally, I would downscale the entire game to fit with your lowest textures but increase the amount of prop clutter and scene complexity. This might be hard to stomach for the perfectionist but a consistent artistic pallete is going to be a lot nicer to look at than a hodge-podge of inconsistency. I'd put greater energy into lighting and composition, and of course how that ties in with level design. Each level should feel like a real place, have a distinctive feel that coincides with the style of gameplay that develops there, and none of this has anything to do with texture resolution.

This is why an old-ass game like CounterStrike still looks better even though its dated technology.

 

Alright did a test, I had to break up alpha for metnaless channel becasuse the AI software does not support alpha channels. But stiched it back together in photoshop. But I get a color difference, dont know if I did something wrong or if the AI changes the colors. I also got a sRGB warning, maybe have something todo with the changed color.

Oh well. Here is the result

 

interesting. are you flipping on an off between the upscaled texture?

what are your thoughts? how easy was it to implement? If does give an impression of a little more detail. Something like if you added a secondary tiling detail map, or just adjusted the textures manually. But was it an easier than doing that? And what of performance? Is this an additional thing operating at runtime?

 

Yeah, once a second I flip. Easy way to visualize difference. It was very easy. Just put the textures in a folder and run a command. About 3-4 seconds per texture to upscale 4 times on a 1080 TI (CUDA). Only problem is that it does not support alpha channel so I had to separate the metalness glossiness into a grayscale image and upscale it seperate and then stitch it back again when done.

 

i tried to find where i read it but couldn't, but i did read once from an environment artist that using alpha channels is bad for performance -- that its much better to use an extra gray-scale image for opacity purposes. I don't remember the details of why but if I come across that again I'll post it.


Overall the difference to me seems negligible. Like, if it's nothihng to do it, sure why not? But I still don't think I would just throw all my textures in there and assume they'll be better. Some textures might not look as good with more contrast and higher frequency detail like that.

 

Obviously upscaling is as good, as good algorithm is.
Which not always may agree with artistic vision of game devs.

 

I have no idea how Unity stores the compiled material. But I dont think its optional how to define the metalness smothness/glossiness, its in the alpha of the Metalnessmap

 

That seems weird, it seems to be hardware related, I think on some old mobile this might be true (due to compression). However having all channel together is only one texture read (get the rgba) while separate alpha mean two texture read, which is an overhead. Then there the whole can of worm with alpha itself due to overdraws and such, maybe it's about separating material from opaque and alpha part.

 

There is no actual alpha in the resulting material. The alpha channel is just used as metadata to describe the smoothness/glossiness of the metalnessmap

 

I think this is relevant

 

yeah.. i dunno about a lot of this. a lot of the magic happens in the "repetitive task." Most AAA games are already basically on the level of fast food, not sure automating out the artist job will make for more artistically focused games. Just gonna be more idle minds with one rich guy to hate.

 

Meh ... will just the same jump from paint canvas to computer art ... remember when design and architecture shop use to hire many people on a single task? Now we the same design and architecture shop are as big but have many people on multiple task. Instead of a single artist spending 5 years on noses for fifa 15, many artist will be devoted to many character and produce higher quality content faster for even bigger, dense and more complex game ... I mean fleshy character with all muscle simulated; so that when the ball hit the face we have all the beautiful skin ripples going through game still derivative though lol, hasn't change in 30 years, tech won't change that

 

heh. alright, you sold me. i want a virtual reality boxing game that i can turn slo-mo on and watch my fist send shockwaves through opponents face.

 

A gif I made to show the developers the difference in white balance. You can also see the detail difference clearly