I was watching some videos of people playing mods on an RTX 3090 and it looked kind of trippy. It looked like a film but something just felt off. Everything was accurate, at least as far as I could tell, but I couldn't tell why I knew it was CG. But after awhile I came to 3 conclusions. Camera movement, repetition and last but not least, unity. When I say unity, I'm talking about the idea that everything is one continuous system. When you go outside, every tree, patch of grass, etc has it's own history. But in games, everything is explicitly placed and nothing looks like it goes together. But at the same time, I understand that things NEED to be this way so the game can function as a game. The player needs to know where to look, what terrain is passable and what terrain is not.
My question is, if we do get every light source is pathtraced at 8K, all nanite meshes at 240fps, etc, etc, then will we ever resolve the things that need to be there but detract from the realism of the experience?
Replies
If you're looking for an experience indistinguishable from reality, VR may get there in 20-30 years.
However, a game that's fun to play is a heavily-designed experience.
There are heavily-designed spaces in real life... theme parks, landscape architecture, japanese zen gardens, etc.
So then it's in the details. Fine resolution. Natural wear. No repetition.
That requires heavy computing power and a lot of memory.
The final 15% will take 85% of the effort.
Hellblade II looked really nice, it's being made with ue5
But it's not only the graphics. It's the animations as well, and the little things that seem off. Even with PBR you cannot get 100% correct lighting, nor with SSS 100% real looking skin. You can't have real time characters with many polygons that adhere to animations and the animations be modular for a gaming experience. It would be ok if it were a film, but if everything's modular to have a gaming experience, it needs like A LOT of polishing and might not even be possible or worth making it
@Larry said:
It's the animations as well, and the little things that seem off.
I was gonna say something like that. Regarding modeling and surfacing I think we're almost there, things like Quixel's Megascans let you take chunks of reality and put them in games, but with animations (or maybe "behaviors") I think there's lots of room for improvement.
A simple example is, in some game, during a serious moment like a mission briefing, say you have a player that keeps jumping around like an idiot, or moving close to a character and staring them in the face. If you do that right now, I'm pretty sure in most games none of the other characters will react, but in a realistic game you'd have them going "What the hell are you doing Steve, back off" and push you away.
I think machine-learning / A.I. will play a big part in bringing innovation to this.
Yes.
I think we'll get there before Eric suggests and it will be driven by ML
Should we? Do we want to?
Not in my opinion
if we keep doing modular shit I dont think it ever will because lif isnt modular 😁
Ikea buildings are tho
For characters as long as we are faking our way around the limitations of using polygon shells with the simplest of deformations I don't think we can. It would take a real paradigm shift to make things interact like real life.
Until then it's like looking at sophisticated versions of
Some of these are pretty good illusions, but only if you're not looking too closely.
I think how convincing it is really depends mostly on the audience.
A lot of people right now cant tell CGI from reality in movies, though it may be obvious to most people here.
yeah no. a human can move like a unreal animation from a video game and it will look real.
why.
cause good renderer.
as always its only the renderer. probably not even the lighting. there are RL scenes which look like no gi but it still looks real. cause good renderer.
It really depends what one means by "looking like real life", because what makes real look real depends on what you're looking at.
Sure, a human can try to move like a videogame character, but it will look really bizarre and uncanny. But we know that's real life, and it's just someone walking weird.
A videogame character moving uncannily triggers us to remember we're playing a game, and notice all the little uncanny things that add up and make it not look quite real.
Renderer be damned, it's about far more than just the renderer— as I learned when I was 15 and obsessed with learning the best renderers, they won't make your work look real! They'll just make it look a bit realer.
Grass and tree models, animations, wind, bounced lighting, color temperature of lights, material properties, realistic shader models, realistic and thoughtfully built models in general (so many prop and environment artists don't understand how real things are built)— all these things need to conspire to make something look real. Else it'll always feel a bit off.
As an example, I think right now something basic, like a white cube in a gray room, could look functionally real already. Or a bottle of aspirin on a wood table.
But something more complex, like a human sitting on the ground, around a campfire in the dense and windy forest, playing the cello, is much harder to get right.
Speak of the devils! Tell me this doesn’t look 3D
yeah man just go to some major US cities and look at the fake as plaster that looks like a repeating normalmap on so many houses
well back to topic, do they have to? thats really personal, but i don't care for anything realistic...
It's theoretically possible.
I think framing it as "look like real life" does not accurately describe the sheer complexity of the problem. (infinite complexity)
We would need the Machine Learning paradigm shift equivalent for game development, and perhaps a trillion times more compute power (making this number up, the idea is we're nowhere near the needed computation).
I don't think the solution would be one that requires manual input from artists or designers or creators. That doesn't scale.
But the illusion of real life doesn't necessarily mean it looks real to everyone. Clearly we can scan some static small environment and light it relatively realistically with limited boundaries so that it looks real to almost everyone.
So it really depends on your definition. Also, the more we limit our scope of the environment, the more feasible it becomes.
I still think that in general that realsitc 3d humans look a little comical, once they are rigged and skiweighted . The more relaistic you 'try' to make them
the more disppointment you are likely to feel.
I am sticking to more stylised stuff for sure. Nahh thats not true I am doing environmental art at the moment
Yes. The more you go into detail the harder you are hit by the uncanny valley. The eye then just searches for it. For me it's more important to have consistent graphics that fits to each other.
Real buildings are modular. All buildings are built from modules - that's what a brick or a roof tile is.
^
they are modular but not snapped on a grid and have exceptions and imperfections and over module are some many hand placed or removed stuff
and for hand made huts made of clays and all different architectures around the worlds at different times?
It is noticeable when you play a map made of modules vs a map made of patch, they feel different, one is expectable
its also part of art directing I guess. realism in proportions stylise and shortcut in proportions etc
If a game doesn't let you do a lot of things, it doesn't matter that it looks completely realistic, it will still feel fake. Like walking around Google Maps, just looking at 3D photographs. It would be novel but not really as exciting as you would think. Complete photorealism would need a high level of interaction with the environment to make it more immersive. However you can have deep immersion without completely realistic graphics. The only limit there is a production schedule. GTAV could let you interact with a lot more and do more things. But creating all that requires time the studio didn't have. If you had a game with the graphics of San Andreas but let you go into almost every building, had tons of interactions with the environment and simulations, then it would be more immersive than a GTAVII that looked absolutely real but you could only steal cars and shoot things.
I think focusing on immersion more than realism is a good strategy, when time affords it.
yeah, i think they will at some point.
Leaving aside augmented reality, I think some games already 'look better than real life'. If you are a tech freak and immediately start analysing how many point lights are in a level, then I don't think games are made for you. A good example would be Boarderlands. I don't like FPS and I don't like Comic style but I love Boarderlands and forgetting all the other aspects that go into making a game what it is, then in this instance give me the shorthand.