Are they ever going to let you customize the population portraits, or at least choose your portrait and gameplay effects separately? It doesn’t make sense to have a unique leader portrait if all of your citizens are of a completely different species.
I don't see that happening here, while generative AI has gotten pretty good at making still images, video and 3D are still way off for now. Maybe in a few years, and as a DLC.
@@galvendorondo I feel like a list of generally appealing ship designs, tabularized, could be used to train a model for combinations of the ship parts. But that would be a huge pain for little gain lmao.
@@KleptomaniacJames it’s not more so about what can be trained, it’s about how the AI will interpret it. If you didn’t know, generative AI does not “draw” like us humans, and similarly, it wouldn’t “sculpt” like us humans too. Adding a 3rd dimension complicates things a lot; add to that the scaling issues and a single point error leading to drastically different outcomes, and well, that technology simply isn’t there yet. 3D AI will take a while to perfect because it needs far more work than 2D, so it’s a ways away.
The generated picture are a bit less accurate on purpose. They separated those 9 results into 3 methods. 3 images are very strict hence pandas in scifi clothes. The next 3 are a bit looser which is where you got raccoons and humans ("pandas" are just a national symbol). Then last 3 is where it has the least restrictions, giving you the robot and the blob monster. This model is specifically fed art from the studio's artists to ensure consistent aesthetics. Frankly it's a genius mechanic. Can't wait to see how it'll evolve in other games in 10 years.
Customization is nice but I rather they use use novel AI technology to craft an interesting and challenging SP opponent. It would solve the big curse of all SP games which is that playing against the computer always becomes very predictable and boring.
Problem is that's a very different kind of AI. You can't really adapt chatGPT, or any other AI for that matter (except maybe previous GalCiv AIs), to play GalCiv4.
@@blitcut9712 yeah not necessarily an LLM although some of the autonomous agents they are working on might be interesting. But stuff like AlphaGo AlphaStar any deep learning technology could potentially be adapted to playing videogames and solving the bad strategy AI problem. It would take some time and effort to develop but could be amazing in the long term if things work out.
@@ares106 While they might be usable there is one major problem which I can think of when applying them to higher level strategy games. The fact that most players don't want the AI to play to win but rather to act as if they're an actual country/civilization inhabiting the world. The most basic requirement of machine learning is the ability to evaluate results, if it can't know if it's doing better or worse it can't learn. But you can't really evaluate if an AI is acting like a country/civilization, at least without a human, meaning that deep learning might be impossible to apply to games like GalCiv.
@@blitcut9712 great point, I do want my AI to play to win, but I would be annoyed if I see AI try to cheese the game like legend of total war or something. The loss function should be tweaked such that the the human player’s satisfaction is the actual reward not necessarily wining the game. This is of course easier said than done as you point out. But I do think it’s crackable considering these days many games report data back to the developers, using this telemetry or even just asking players for their satisfaction after a game could be a tool harnessed in improving a deep learning AI.
@@ares106 forza motorsport is using machine learning but its pretty expensive and they have been building it up for a decade its a ways out before its fully adopted its not easy you need people with the know how plus these games are complex even for the average human were still a decade away from this but thats not saying we cant get good ai now
What I really really dont understand about this franchise is the system requirements. The game has basically been the same since part 2, not many graphic improvements, not a huge increase in the ingame-systems, (im sure stellaris has a LOT more to calculate) yet somehow the system requirements were high with part 2 and continue to get higher. I just dont get it.
@@davidptNo youre right, because Im not a game designer, but I have been playing games for 30 years including all of these Galciv games. So yeah,, I have an opinion on the matter. Goodluck designing the next RDR2 and Elder Scrolls btw, MR expert game designer in the youtube comment section.
@@kearseymorton2078 Well at least you had very interesting things to add to the discussion and youre not one of those pathetic people just trying to look smart in youtube replies.
Wait where is this thing pulling the art assets for for it's image creation? Is it their own model that they've trained or are they using the work of scraped, unpaid, uncredited artists like so many of these things do? Kudos if it's their own model, but some of those images look a lot like composited Rocket Raccoon fan art.
@@Thortacozone Yeah I followed up and watched some of their dev log videos where they talk about using exclusively their own assets Honestly this a cool and responsible use for the tech, good on them.
I would love a full review covering the different aspects of the game in detail
Are they ever going to let you customize the population portraits, or at least choose your portrait and gameplay effects separately? It doesn’t make sense to have a unique leader portrait if all of your citizens are of a completely different species.
Would be great to integrate a way for AI to make ships based on prompts
I don't see that happening here, while generative AI has gotten pretty good at making still images, video and 3D are still way off for now. Maybe in a few years, and as a DLC.
@@galvendorondo I feel like a list of generally appealing ship designs, tabularized, could be used to train a model for combinations of the ship parts. But that would be a huge pain for little gain lmao.
@@KleptomaniacJames it’s not more so about what can be trained, it’s about how the AI will interpret it. If you didn’t know, generative AI does not “draw” like us humans, and similarly, it wouldn’t “sculpt” like us humans too. Adding a 3rd dimension complicates things a lot; add to that the scaling issues and a single point error leading to drastically different outcomes, and well, that technology simply isn’t there yet. 3D AI will take a while to perfect because it needs far more work than 2D, so it’s a ways away.
The generated picture are a bit less accurate on purpose. They separated those 9 results into 3 methods. 3 images are very strict hence pandas in scifi clothes. The next 3 are a bit looser which is where you got raccoons and humans ("pandas" are just a national symbol). Then last 3 is where it has the least restrictions, giving you the robot and the blob monster. This model is specifically fed art from the studio's artists to ensure consistent aesthetics. Frankly it's a genius mechanic. Can't wait to see how it'll evolve in other games in 10 years.
I've been trying to get my brain around this one but it looks great
Customization is nice but I rather they use use novel AI technology to craft an interesting and challenging SP opponent. It would solve the big curse of all SP games which is that playing against the computer always becomes very predictable and boring.
Problem is that's a very different kind of AI. You can't really adapt chatGPT, or any other AI for that matter (except maybe previous GalCiv AIs), to play GalCiv4.
@@blitcut9712 yeah not necessarily an LLM although some of the autonomous agents they are working on might be interesting. But stuff like AlphaGo AlphaStar any deep learning technology could potentially be adapted to playing videogames and solving the bad strategy AI problem. It would take some time and effort to develop but could be amazing in the long term if things work out.
@@ares106 While they might be usable there is one major problem which I can think of when applying them to higher level strategy games. The fact that most players don't want the AI to play to win but rather to act as if they're an actual country/civilization inhabiting the world. The most basic requirement of machine learning is the ability to evaluate results, if it can't know if it's doing better or worse it can't learn. But you can't really evaluate if an AI is acting like a country/civilization, at least without a human, meaning that deep learning might be impossible to apply to games like GalCiv.
@@blitcut9712 great point, I do want my AI to play to win, but I would be annoyed if I see AI try to cheese the game like legend of total war or something. The loss function should be tweaked such that the the human player’s satisfaction is the actual reward not necessarily wining the game.
This is of course easier said than done as you point out. But I do think it’s crackable considering these days many games report data back to the developers, using this telemetry or even just asking players for their satisfaction after a game could be a tool harnessed in improving a deep learning AI.
@@ares106 forza motorsport is using machine learning but its pretty expensive and they have been building it up for a decade its a ways out before its fully adopted its not easy you need people with the know how plus these games are complex even for the average human were still a decade away from this but thats not saying we cant get good ai now
say goodbye to artists
What I really really dont understand about this franchise is the system requirements. The game has basically been the same since part 2, not many graphic improvements, not a huge increase in the ingame-systems, (im sure stellaris has a LOT more to calculate) yet somehow the system requirements were high with part 2 and continue to get higher. I just dont get it.
Well, that's basically what you said is entirely wrong! That's why you don't get it. You don't know game design at all
@@davidptNo youre right, because Im not a game designer, but I have been playing games for 30 years including all of these Galciv games. So yeah,, I have an opinion on the matter. Goodluck designing the next RDR2 and Elder Scrolls btw, MR expert game designer in the youtube comment section.
Then maybe enlighten him instead of going 🤓@@davidpt
@@playerone7663
as usual: "i have been playing games for 30 years" + "I have an opinion" = huge amount of BS
@@kearseymorton2078 Well at least you had very interesting things to add to the discussion and youre not one of those pathetic people just trying to look smart in youtube replies.
Totalitarianism/Nihilism 🗿
Wait where is this thing pulling the art assets for for it's image creation? Is it their own model that they've trained or are they using the work of scraped, unpaid, uncredited artists like so many of these things do? Kudos if it's their own model, but some of those images look a lot like composited Rocket Raccoon fan art.
its based on stardocks 30 years of art and they are hiring artist to make more art they can use for the ai
@@Thortacozone Yeah I followed up and watched some of their dev log videos where they talk about using exclusively their own assets Honestly this a cool and responsible use for the tech, good on them.
That also explains why overall feel of the images fits in with those of the official factions.
They should have used stable diffusion or something with higher database.
Would be a perfect game if not for the so called battles