3.80 star(s) 64 Votes

Nihil5320

Member
Jul 2, 2022
365
925
yea pretty much. Man that game sounded so good the first time I found it but GOD it was so shit.
The fucking building components wouldn't fit together properly no matter what you did lol, and there are certain areas of the maps where FPS drops by like 70% for absolutely no reason at all.
Nevermind the extremely infuriating moans that drove me fucking nuts.
Oh yea and none of the girls' mechanics worked properly. I think I stopped playing because the one girl I had just slept all daay and I couldn't stop her no matter how full her stats were XD Absolute trash of a game.
The concept was pretty good but yeah the implementation was about what you would expect. AI in games has something of a tendency to turn into spaghetti code so you either end up with bugs or very limited and basic behaviour. Or both.

I'd love to see a game built on the same concept (even a non-porn game) that really went all out. Epic introduced learning agents into Unreal recently too which should make this easier, so instead of using behaviour trees you could train a DNN based model to optimize for certain goals with reinforcement learning. You could do this before but this definitely lowers the bar to entry, although it still comes with downsides like the need to retrain models and difficulties in modifying/explaining behaviours.

Combine it with stuff like conversational AI and physics based animation and you could end up with, at the very least, a fairly awesome (if janky) tech demo. The trick there would be mapping the LLMs output to actual in game actions and coming up with a reward function that can get your behaviour model to pick up basic actions then develop them into achieving more complex goals.

Kind of makes me wish I'd done game dev as it would be an amazing project, plus it would be fucking hilarious if the first title to actually implement some of these approaches was a porn game.
 

HardcoreCuddler

Engaged Member
Aug 4, 2020
2,543
3,243
The concept was pretty good but yeah the implementation was about what you would expect. AI in games has something of a tendency to turn into spaghetti code so you either end up with bugs or very limited and basic behaviour. Or both.

I'd love to see a game built on the same concept (even a non-porn game) that really went all out. Epic introduced learning agents into Unreal recently too which should make this easier, so instead of using behaviour trees you could train a DNN based model to optimize for certain goals with reinforcement learning. You could do this before but this definitely lowers the bar to entry, although it still comes with downsides like the need to retrain models and difficulties in modifying/explaining behaviours.

Combine it with stuff like conversational AI and physics based animation and you could end up with, at the very least, a fairly awesome (if janky) tech demo. The trick there would be mapping the LLMs output to actual in game actions and coming up with a reward function that can get your behaviour model to pick up basic actions then develop them into achieving more complex goals.

Kind of makes me wish I'd done game dev as it would be an amazing project, plus it would be fucking hilarious if the first title to actually implement some of these approaches was a porn game.
Actual AI AI would be interesting in games but I feel like that still far off...AI's are barely able to respond to extremely specific requests right not, let alone being able to interact with a player which may want a thousand things.
I really believe decision / behaviour trees are still the way to go and can be implemented properly with some care and proper separation, but it takes more skill than is usually available to do that...People that can implement that kind of thing cleanly don't seem to be working game dev jobs unfortunately...
 
  • Like
Reactions: BarnMTB

Nihil5320

Member
Jul 2, 2022
365
925
Actual AI AI would be interesting in games but I feel like that still far off...AI's are barely able to respond to extremely specific requests right not, let alone being able to interact with a player which may want a thousand things.
I really believe decision / behaviour trees are still the way to go and can be implemented properly with some care and proper separation, but it takes more skill than is usually available to do that...People that can implement that kind of thing cleanly don't seem to be working game dev jobs unfortunately...
It's not as far off as you think. Getting an agent to respond to specific requests is a bit tricky, as mentioned above matching an LLMs output to actual actions is not always straight forward, but you can mitigate this to some extent by very strictly limiting the dialogue flow and available actions. for example is a service recently announced for exactly this and their dialogue manager will do stuff like trigger new quests based on keywords, which is a fairly basic approach, but the application could additionally run sentiment analysis on the output (and/or user input) to only trigger events in a given context.

It's the same with behaviour trees. You aren't going to get a completely free ranging agent capable of doing absolutely anything in an environment but one making higher level decisions like what to do or where to move to? That's arguably easier to achieve with ML these days than a complex behaviour tree, and even in environments with a very large action space you can use parameterized actions or action masking to keep things reasonable. Setting up the training environment and reward function is the hard part but once you've got that the model can figure out all the really finicky bits.

Finally layer on to generate animation dynamically in situations that require it, perhaps dodging items or traversing obstacles, and you can end up with a pretty convincing facsimile of a genuinely intelligent agent even if it is perhaps more limited than the sci-fi fever dreams some are expecting when they see the word "AI."

That said I work with ML, in fintech specifically, and your last comment is probably on point. This is changing though as more and more universities introduce machine learning modules, better and easier to use libraries become available etc. Obviously we almost certainly won't see this stuff first in porn games but it wouldn't surprise me if indie devs are quicker to test it out than larger firms.
 

SortaLewd

Member
Feb 26, 2019
375
511
yea pretty much. Man that game sounded so good the first time I found it but GOD it was so shit.
The fucking building components wouldn't fit together properly no matter what you did lol, and there are certain areas of the maps where FPS drops by like 70% for absolutely no reason at all.
Nevermind the extremely infuriating moans that drove me fucking nuts.
Oh yea and none of the girls' mechanics worked properly. I think I stopped playing because the one girl I had just slept all daay and I couldn't stop her no matter how full her stats were XD Absolute trash of a game.
It's not so bad modded, but that's usually how Illusion games are in general. Base game is okay, modded is superior.
 

monomulti

Member
Jan 13, 2018
115
105
They are clearly concerned about the lack of subscriber growth.
It's pretty pathetic that their primary concern at this point is still subscriber growth. They've had minimum 14k+ monthly subs for over 2 years now. Several millions of dollars in funding and still no full release, previews of any actual content, or even a roadmap. Only a tech demo that has hardly any more content than 0.34 released 2 years ago.
Somehow all that money wasn't enough it seems. I'm certain they would just abandon the game if the current trend continues and sub counts tank. I guess we're supposed to fund Project Helius's entire development cost until they milk out even more profit post-release, if this game ever releases.
 

HardcoreCuddler

Engaged Member
Aug 4, 2020
2,543
3,243
It's not so bad modded, but that's usually how Illusion games are in general. Base game is okay, modded is superior.
I mean yea, but are there mods / plugins / whatever that make the actual game part of AI any better? (ignoring the studio, that's basicaly the same as HS2 afaik)
I myself don't know any mods that make AI's game better but I'd like to know if you do, because I may just give it another crack
 
  • Like
Reactions: BarnMTB

benisfug

Member
Aug 18, 2018
303
865
it always drops on payday for everyone, no?
If you look at their graph they used to have monthly growth to offset the drop. It was trending down but only by a couple hundred month to month. Now there's no growth, it's just a flat line and a straight drop of 1k-ish subs.
 
  • Like
Reactions: BarnMTB

SortaLewd

Member
Feb 26, 2019
375
511
I mean yea, but are there mods / plugins / whatever that make the actual game part of AI any better? (ignoring the studio, that's basicaly the same as HS2 afaik)
I myself don't know any mods that make AI's game better but I'd like to know if you do, because I may just give it another crack
Last I had AI Shoujo installed I was just using the repack IIRC. There were some mods included in the repack that made the building a little more enjoyable like removing building placement restrictions, free form building or snapping things into place(iirc), and lessening material costs/removing material costs. There were other mods that made personalities more varied IIRC, but none that drastically changed gameplay. I know there was a mod that add survival elements with the girls needing to eat/drink/sleep, pregancy(the girl temporarily leaves the island after being pregnant for some time), but other than that not too many gameplay altering mods I remember atm. I remember I just turned on the survival mod and increased the number of girls that could be spawned at once and just played it like a management game. Building a house with interactable stuff for the girls to mess around with or to interact with each other and making food sources, that kind of thing.
 
Last edited:

HardcoreCuddler

Engaged Member
Aug 4, 2020
2,543
3,243
Last I had AI Shoujo installed I was just using the repack IIRC. There were some mods included in the repack that made the building a little more enjoyable like removing building placement restrictions, free form building or snapping things into place(iirc), and lessening material costs/removing material costs. There were other mods that made personalities more varied IIRC, but none that drastically changed gameplay. I know there was a mod that add survival elements with the girls needing to eat/drink/sleep, pregancy(the girl temporarily leaves the island after being pregnant for some time), but other than that not too many gameplay altering mods I remember atm. I remember I just turned on the survival mod and increased the number of girls that could be spawned at once and just played it like a management game. Building a house with intractable stuff for the girls to mess around with or to interact with each other and making food sources, that kind of thing.
Ah okay, I had that stuff as well. Wasn't too enjoyable to me though...I only really wanted one single girl and she was just broken all the time. I remember being able to free place building blocks as well but my issue was they never fit together properly lol As in I couldn't get them to look good at all
 

Nihil5320

Member
Jul 2, 2022
365
925
kinda crazy, how we got string alone for what? 6+ year? with Sketches and shit, not even one single screenshot of the so call campaign and still 13k sub
Their Patreon launched September 2016, literally the instant Patreon changed their community guidelines to allow adult content, so nearly seven years.

That's not including the fact that it was already in development before the Patreon launch so realistically it's probably closer to eight or nine years if you go from the first previews of "Daisy in Summer." That fits with the timeline of them starting pretty much the moment Epic launched UE4 with a public license too.

If we take their claims at face value that means 9 years for a very limited and buggy tech demo. According to Helius they've completed the script, voice acting, level design, ux design and mocap for campaign multiple times but just... never released anything.

It's what makes this project so fascinating. They can do the absolute minimum to create a functional tech demo, make bold faced lies repeatedly and thanks to the design of sites like Patreon they can keep it going for years as the developers themselves have absolute power to hide feedback, there's no oversight of their marketing and they can even delete their prior statements if they want to walk stuff back.
 

jenrenn

New Member
May 7, 2023
1
0
kirito mod says failed opening file logs core and its unable to start. apparently it doesnt exist. any solution?
 
3.80 star(s) 64 Votes