It's currently on hold due market conditions and technology limitations (not many interested in a visual novel with AI generated visuals). The sequel to this is almost complete though, it uses traditional graphics: https://deepglugs.itch.io/euryales-gambit.
I agree. I'm putting this as an optional next project after Euryale's Gambit. Patreon's will vote for either this or two others. Also, I have been experimenting over the last month with some new AI tech. Unsure if it'll be better, but if it is... could make this game a lot more doable even if it doesn't get the votes.
i have some feedback that i hope you will take onboard and inprove the V.N.
i'm not sure if if its the encoding or size compression but on a 24" the image looks like pixel vomit is there something that can be done to make it much more clearer and sharper image even if it mean make the file size much larger it would be a huge blessing
it's not the compression. It's a consequence of the AI image generator starting with a 256px image and eventually upscaling to 1024px. It's better quality than just resizing, but it's not perfect as you can see.
The issue in starting with larger images is the amount of GPU memory required to train the AI image generator. In addition, there just aren't AI technology that can produce great quality images at higher resolutions. I'm keeping my eye out for the tech to be developed. I'd love to continue developing this VN.
Going to go with the crowd here, story seems like it would be funny and interesting, but the images throw the entire thing off. If the developer cleans up the images or really redoes them, it would be 5 stars. I had to stop after 4 minutes in.
They aren't broken, it's just that our AI artist isn't very good at drawing yet. She's learning how to be better. It looks like version 0.3's images will still be a bit blurry. Hopefully this will improve with time. How did you like the story so far?
there were cases where the A.I. drawings were able to distingish some features but then there were other cases where the A.I. program in charge spat back a smudge that if I wasn't paying attention I would have no clue who was talking or things like that. A simple solution if you are using a learning A.I. would be have it analize pictures of real people from the internet and continued practice drawing figures it may also help if you take the time every once in a while to check the progress and manually correct the A.I. like you would with a kid who is learning a new skill, a few minutes of it drawing then picking the best one would help it recognize what is expected of the program and in cases where there is no best one, hand correction may be needed. just a few programming tips when it comes to learning A.I.s
You are absolutely right. We mentioned in our last devlog that 0.5 will do exactly that. We will be expanding our training dataset to include helpers for specific scenes and trying out new AI network architectures and experimenting some more. Thanks for the feedback.
Haha. I certainly couldn't say "succubi" at 2 months old, but yes, "daturanet" has some room for improvement. We are working on it and it should be improved by 0.2 if not 0.3. Also, thank you for the review!
← Return to game
Comments
Log in with itch.io to leave a comment.
is this game still alive? it's been awhile since the last update
It's currently on hold due market conditions and technology limitations (not many interested in a visual novel with AI generated visuals). The sequel to this is almost complete though, it uses traditional graphics: https://deepglugs.itch.io/euryales-gambit.
that's a shame it looks like it has some good potential, hopefully, it doesn't get abandoned, but I'd understand if it did.
I agree. I'm putting this as an optional next project after Euryale's Gambit. Patreon's will vote for either this or two others. Also, I have been experimenting over the last month with some new AI tech. Unsure if it'll be better, but if it is... could make this game a lot more doable even if it doesn't get the votes.
i have some feedback that i hope you will take onboard and inprove the V.N.
i'm not sure if if its the encoding or size compression but on a 24" the image looks like pixel vomit is there something that can be done to make it much more clearer and sharper image even if it mean make the file size much larger it would be a huge blessing
it's not the compression. It's a consequence of the AI image generator starting with a 256px image and eventually upscaling to 1024px. It's better quality than just resizing, but it's not perfect as you can see.
The issue in starting with larger images is the amount of GPU memory required to train the AI image generator. In addition, there just aren't AI technology that can produce great quality images at higher resolutions. I'm keeping my eye out for the tech to be developed. I'd love to continue developing this VN.
Thanks for the feedback! I appreciate it.
ok i understand and the high end GPU's going for stupid £££ i can understand.
improving A.I in-game is very interesting. voiced characters adds quality into the game. Very promising!
Going to go with the crowd here, story seems like it would be funny and interesting, but the images throw the entire thing off. If the developer cleans up the images or really redoes them, it would be 5 stars. I had to stop after 4 minutes in.
hey i just wanna know, is the images broken or something? because it seems like blury, sorry for my bad english...
They aren't broken, it's just that our AI artist isn't very good at drawing yet. She's learning how to be better. It looks like version 0.3's images will still be a bit blurry. Hopefully this will improve with time. How did you like the story so far?
there were cases where the A.I. drawings were able to distingish some features but then there were other cases where the A.I. program in charge spat back a smudge that if I wasn't paying attention I would have no clue who was talking or things like that. A simple solution if you are using a learning A.I. would be have it analize pictures of real people from the internet and continued practice drawing figures it may also help if you take the time every once in a while to check the progress and manually correct the A.I. like you would with a kid who is learning a new skill, a few minutes of it drawing then picking the best one would help it recognize what is expected of the program and in cases where there is no best one, hand correction may be needed. just a few programming tips when it comes to learning A.I.s
You are absolutely right. We mentioned in our last devlog that 0.5 will do exactly that. We will be expanding our training dataset to include helpers for specific scenes and trying out new AI network architectures and experimenting some more. Thanks for the feedback.
I will forever refer to succubi as succus from now on.
Haha. I certainly couldn't say "succubi" at 2 months old, but yes, "daturanet" has some room for improvement. We are working on it and it should be improved by 0.2 if not 0.3. Also, thank you for the review!