Decline Are we ever going to see tech that can produce cheaper game assets?

[Bohrain]

Just a random thought. Obviously developing a game requires a minimal amount of know-how these days with commercially available engines and asset stores compared to the time when everything had to be coded from scratch with machine languages. And some of them machine learning doodads can upscale low-resolution textures with varying amounts of success. Yet those ready-made options also teach the audience to develop a distaste for the default mechanics and assets, whether it's RPGMaker tilesets or Unity asset store models. So while it is technically easier to produce a functional game with less skill, it's for the most part meaningless for less skilled developers and smaller teams since the customers don't want to touch what they see as the standard engine derivative.

But are we ever going to see something like an algorithm being able to produce portraits of the same character from different angles and variety of expressions? Or something like music or automatic animation rigging? Some people might say that AI Dungeon counts for text adventures, but it feels like a novelty gadget that can't really produce something like a three act story or clear character arcs which is what you want to see in a storyfag title.

 

[Takamori]

I could see that happening, every engine having a discount bucket store of assets that can use it freely on their games after paying for it. A way for artists to make a quick buck.
The consequence is having several games with the same asset and being souless thanks to that. Oh I saw that monster before from Generic game that I tried previously and so on, I think its a reality already in the indie garbage bin.

 

[Morpheus Kitami]

There are already tools that produce 2D tilesets that look relatively acceptable. And there's also tools that produce acceptable music, and if not, someone can just adapt classical music or whatever. Sound too, well, except the real stuff. (you'd use a library for that anyway) I'm sure there are tools for the rest of one's 2D graphical needs. I couldn't tell you how far off the character portrait thing is, but automatic animation rigging seems really, really far off in technology.
But none of that matters, because the usual people who make generic RPGMaker titles are not going to do anything with it. Let me ask you, are you aware how much free sound, music and graphical assets are around right now? As in you can legally use them for free in a commercial game. There's even more if you want to search through public domain assets. People who want to make their game not look like a generic game will do so, and the people who just want to make RPGMaker Game #9999 will do so.

 

[cpmartins]

AI is an interesting subject, I've studied a bit on my compsci days. Every teacher/researcher I had spoken to said the same thing: As long as complete parallel processing is unavailable, True AI (the kind that can produce useful creative output for human use) would be almost impossible. That means having a neural network composed of billions of discrete processing units and a
communication medium that won't be slowed by them when they switch their internal status to emulate another set of engrams (context switching).

The biggest bottleneck at the time was mass deadlocking, not sure about the situation with AI specialized hardware today since I haven't been following it for a long time.

 

[Excommunicator]

I don't want it. Cheap and easy assets will make it easier for business savvy people flood the market with bad products to make a buck. It isn't going to release the shackles from a horde of creative types and allow them to suddenly make great games.

Anyone who has a good enough game or concept ready to go and the talent to achieve it will be able to source the resources to make it happen. Those who are passionate about making it happen will make it happen whatever barriers to content production might exist.

 

[Tavernking]

Already begun. I forget the name of the website but it changes photographs of faces into really nice portraits of similar quality to Pillars of Eternity for example. The tech was a bit dodgy, but in a few years I can see myself using it for RPG character portraits if I make a game that needs them.

 

[Nifft Batuff]

They already using similar techniques even if they don't tell you. If they aren't using them it is only due to the fact that human work is cheaper in some countries.

 

[a Goat]

We're going to sir tech

 

[Luckmann]

But are we ever going to see something like an algorithm being able to produce portraits of the same character from different angles and variety of expressions?

The short answer to your thread question is "Yes", but this, specifically, is something we already have the technology to do. You can with relative ease feed an algorithm examples of art and then tell it to turn a photograph into that art style. There is little preventing you from, for example, have it look at your 3D-based avatar in a game, fill in potential blanks, and spit out an "artistic" rendition from any angle.

Now, I'm not saying this is going to make it into games next year or anything, and currently, even with art from tens of thousands of art pieces, it still comes out looking like shit sometimes, but these things are constantly getting better.

And we're also at a point where you can reliably imitate voices, like audio deepfakes, once you've fed it someone's voice as a sample. You could conceivably create "databanks" of "voice templates" by having people read a lengthy text and a number of specific words, and then have the AI generate near-pitch-perfect renditions of custom text.

I expect things like this to start making their way into AAA titles in the next decade or so.

Edit: One thing I saw a while ago was an "AI" that was being trained in extrapolating 3D assets based on 2D photography. This will likely never be reliably perfected (I believe), but it's a pretty cool idea. Theoretically in the future, you can snap a picture of something and the system will shit out a usable 3D asset, or at least a partial that you can work with.

 

[Lurker47]

Sure.

 

[Butter]

Sure, but Vogel will still find reasons to make games that look like they came out in 1994.

 

[Luckmann]

Sure, but Vogel will still find reasons to make games that look like they came out in 1994.

I wonder how a game would come out looking if we asked an "AI" to make the art for it, but we only fed it all games 1990-1999.

 

[Kruno]

Sure, but Vogel will still find reasons to make games that look like they came out in 1994.

Those actually look better than the ones he makes today.
The old school Exile graphics still look good today.

 

[Darkzone]

Already begun. I forget the name of the website but it changes photographs of faces into really nice portraits of similar quality to Pillars of Eternity for example. The tech was a bit dodgy, but in a few years I can see myself using it for RPG character portraits if I make a game that needs them.

Not only this. Photogrammetry for static assets. Automatic rigging and weight painting. etc.. etc...

They already using similar techniques even if they don't tell you. If they aren't using them it is only due to the fact that human work is cheaper in some countries.

There is nothing so cheap than just to press a button.

Edit: One thing I saw a while ago was an AI that was being trained in extrapolating 3D assets based on 2D photography. This will likely never be reliably perfected (I believe), but it's a pretty cool idea. Theoretically in the future, you can snap a picture of something and the system will shit out a usable 3D asset, or at least a partial that you can work with.

It is already better and more representative of the real object than the work of an artists.

 

[DalekFlay]

Pretty sure that fancy PS5 demo that Epic cooked up was said to have technology in it that turned pictures into in-game textures, similar to things people mentioned above. Doing it at that kind of AAA scale is impressive. Unless I'm totally misremembering what they said.

 

[Darkzone]

Pretty sure that fancy PS5 demo that Epic cooked up was said to have technology in it that turned pictures into in-game textures, similar to things people mentioned above. Doing it at that kind of AAA scale is impressive. Unless I'm totally misremembering what they said.

You mean the Unreal Engine 5 demo in PS5 (works also lower specs machines, but not in this resolution and framerate). And yes this are photogrammetry assets (most of them), besides the mesh of the female protagonist.

 

[rusty_shackleford]

That tech is constantly being made.
e.g., Allegorithmic's substance designer & painter is a massive improvement over the previous workflows

 

[Gregz]

 

[Stormcrowfleet]

IDK about AI making everything itself. But computer hardware and technological tools already help people make much better asset in absolutely bunker times developping work processes that are super efficient. Case in point: that chinese dude that solo developped an FPS that looks like a AAA. He developped his own tools, notably for facial recognition, that saved him tons of work time. The same project many years ago as a solo dev would have been either impossible or take 10 years. But the guy is intelligent and we have access to new tools now.

 

[rusty_shackleford]

Photogrammetry requires a lot of manual touchup afterwards, and nothing it produces it suitable for animating without a full retopo.

 

[Mr. Magniloquent]

cpmartins I think the biggest impediment to AI is that we are still coding in binary languages on binary circuits. We aren't going to get AI capable of real creative ability until we have mature quantum computing. In college I had a compsci buddy who was big into neural networking. We debated about the potentials of AI for years, but he eventually agreed.

 

[Luckmann]

cpmartins I think the biggest impediment to AI is that we are still coding in binary languages on binary circuits. We aren't going to get AI capable of real creative ability until we have mature quantum computing. In college I had a compsci buddy who was big into neural networking. We debated about the potentials of AI for years, but he eventually agreed.

The concept of "AI" has unfortunately come to be synonymous with "machine learning", which is what is actually being discussed. True AI isn't realistic for a long time, and would likely require quantum computing, but one issue is that we don't even know what we need. We still don't know what produces functional consciousness or why, so we cannot effectively replicate it even theoretically. For all we know, the distinction may not even be relevant.

 

[Gregz]

Photogrammetry requires a lot of manual touchup afterwards, and nothing it produces it suitable for animating without a full retopo.

I can't imagine algorithmics won't grow around this technology to resolve the problem.

It seems incredibly promising.

 

[Darkzone]

Photogrammetry requires a lot of manual touchup afterwards, and nothing it produces it suitable for animating without a full retopo.

Photogrammetry is just a descriptive name / description for technologies (not just one) of getting information by measuring from a set (not 0) of pictures about a object or the environment. The result is dependend on the algorithms within a software that uses it, then the photos quality and amount and etc. (More pictures from differerent angles are always better.) And while it still needs many times a clean up it is just a matter of time until this is also fully automaticised. How the mesh is created from the photos is also a matter of the software and its algorithms. A friend of mine uses a body mesh that was created from pictures and it had to be cleaned up, because the software connected the 3d points ( from the cloud) wrong, but not because 3d points were measured wrong (Photogrammetry).
Example of problems due to few pictures in Reality Capture (2018):

Spoiler

Photogrammetry requires a lot of manual touchup afterwards, and nothing it produces it suitable for animating without a full retopo.

I can't imagine algorithmics won't grow around this technology to resolve the problem. It seems incredibly promising.

If you would have witnessed the development over the last few years then would think that this will be solved within the next 2-10 years, with a good expectation of 6 years.
Also most of the current in use "commercial" Photogrammetry software came out around 2013 - 2014, but naturally they are constantly updated.

So for any one who wants to try it out i recommend the free private version of 3DF Zephyr, but Meshroom is completely free and open source.

 

[Darkzone]

My own quote "It is already better and more representative of the real object than the work of an artists." is a bit trobbeling for me, because better is a statement of preference without stating the measured attribute. What i mean it that it is more realistic since it can display what was exactly seen (or even not seen with the human eyes) than any similar 3d object that an artist can create in a reasonable time. Example tree stump in Reality Capture (2019) made from a video:

Spoiler

But the previous example is only possible with a when the quality and the quantity are given. Therefore i want to point here also out is that the quality and quantity of the pictures for photogrammetry is currently still very important (2018):

Spoiler

But even the quality will lose a bit of the importance, because Data that is distributed throughout a video in different frames can be brought together for each of the individual frames to raise their quality for the photogrammetry. "Grandmother Despina" is an example of how the quality can be raised by simple measures like upscaling.

Headshot pluging for Character Creator 3 by Relusion is creating face models from one picture:

Spoiler