I like your stuff so much that I am going to withhold myself from checking on this project, just so that I can experience a larger experience. I will hold back until October at most
I hope you'll enjoy it! I hope you know you won't be put off by the artstyle though, I'd hate for you to hold yourself back only to not enjoy it when you do actually play.
This is seriously very impressive. Most AI games are not that great, but this one is amazing, keep up the good work
Glad you liked it!
Really I've found that if you put in time and effort, it isn't hard to get good-looking AI art. SDXL, specifically autismmix, makes generating porn extremely approachable.
v2 is gonna be both furry and non furry?
Probably not. I've automated a lot of the most tedious parts but most of the final quality comes from checking and polishing the images manually. I'd probably say the non-furry version of v2's content would come a week or two afterwards.
wow this looks really good. two different versions as well its like two games in one.
Yeah, I thought about making it an entirely separate game, but I committed to making the MC a fluff maniac so rewriting would be annoying.
where do you use the cheat codes listed on the first page?
You use them in Angelica's office, they're referred to as postal codes.
Furry art is good, but the new shortstack stuff is Top Tier, especially for Ai, respect for trying to do it as ethically as possible.
Shortstacks are underutilized. More games need big booty shortstacks!
I will say though that it isn't top-tier in my eyes yet. Lots of inconsistencies in the non-furry version drive me crazy. Cayenne's hair in particular...
But these inconsistencies are mostly due to the workflow I'm using. I'm training a model based on the best of the game's images to hopefully improve the quality even further!
I dont think the use of AI is ethical. did you ask the artist for permission even to have an model trained on them?
I wanna understand your point of view abit more before I really start to argue over how the AI model was made
You raise a good point, though saying "the" artist implies that the syuro lora is doing most of the work.
Even if I did have Syuro's permission (which, to be clear, I do not, I don't want to mislead you there), I would still be piggybacking off of all the other artists who's works were used to train the Easyfluff model.
To simplify it, let's say for each image it generates the AI will draw upon bits of random images in the training data. This means hypthetically nearly every part of the final image could be taken from works by Jay Naylor, (who I strongly believe was in the training data). The LoRa comes in on the final step, like a coat of paint on a finished house.
The while situation is extremely complex. Especially when you consider that many AI models got their start when an insider leaked NovelAI's proprietary training data. Nearly every level of the technology's development is conparable to piracy, which is already a morally grey area.
What I really wish is that I could make a list of the artists used to train easyfluff so I could include the entire list of them. Easyfluff in particular is the most rigid model I've ever used, meaning that most of the visual style comes from the model instead of the lora.
Hm.... While the Non-Furry version is good, the dialogue kinda conflicts at time. Then again, this is a rough alternate version of the Furry version of the game so I'll give it a pass.
The non-furry version of the game has some odd choices.
No ears and tails, but Riley still has a red rocket.
Aside from that, fun game. Hope it gets as many male characters as Hentai University.
Fun fact, when developing the non-furry version at the start I wanted to still include the character's ears and tails. That would have been a nice middle ground, but the way AI generation works is that often elements will bleed into each other.
Basically, if every image in the training data that has a sunny sky also has clouds then whenever you ask for a sunny sky, you're likely also to get clouds. There are ways around this, but they rely on how the model was trained.
Asking for details like tails and animal ears would usually give animal noses as well, so I only copied the furry character's elements that didn't cause problems.
As another fun fact, this element bleeding actually caused problems for the furry version too. Mary-Lou doesn't have a nose because on the first model I used the bovine tag bled into horribly ugly snouts.
I'd like to do more shortstack games in the future that don't have these narrative inconsistencies, mostly because there aren't enough big-bottom'd goblin girls in the world.