Here is the thing I have been talking about, albeit its not the VN with a LLM gearing somewhere behind the scenes.
2 Likes
I’m pretty sure it is actually possible to plug open AI’s API into renpy. It is running in python afterall. The issue is that I don’t think the technology is really there. GPT 3.5’s context memory is only a few thousand tokens and it wouldn’t be much of a game about weight gain if the characters forgot they were fat every few minutes.
That and the API costs aren’t exactly cheap. It’s a cool idea though.I just don’t think it would work here. Maybe a different game when this one is fully cooked.
1 Like
Just playing through intro sequence:
-
00254-453073610.pngneeds to be renamedbg intro twelve.pngingame/images/intro - Some of the images have what looks like either compression artifacts or reduced colorspace, most noticeable with purple bleeding into blue bleeding into blue on clothing / hair / etc.
- The AI rendered artwork is definitely a departure. I know this one’s been beaten to death already but I haven’t figured out how I feel about it yet; probably need to play more. I did already notice inconsistencies in eye color / roundness of cheeks in the intro sequence shots alone, things that stable diffusion is often inconsistent about.
Great update. I enjoyed the new shrine content. I think the stable diffusion is not a bad choice, but I would go over the graphics manually a few times. For the mirror images I noticed some black faded patches on the arms and face as well as a few double belly buttons. On others, the worse offender was a misplaced butt hole on Chika in a image.
Hope you are able to resolve the house issue you mentioned and I look forward to the next release.
Tell me more about this rogue butthole lmao.
I’ll get rid of it.
Maybe when we’ll have a gpt 4.5 or optimized local models for that, yes. Vector databases and semantic banks are a thing, but it’s too much work for now.
Non-ironically, it’s a skill issue. I don’t consider myself in a position to teach and poke someone around, but with decent models and loras it’s possible. And looks good.
Models like:
1 Like
The emotional roller coaster I went on upon seeing a new update, followed by the words “AI Art”
I have very mixed feelings over this, and I’m sorry to say I don’t think i’ll be enjoying the game as much with this inclusion. I respect your choices, your game should be made how you want; However, the new images remove a lot of the charm the old models had. Replacing them with the same artstyle as most other AI generated works.
So I will simply wish you good luck on any and all future updates, cheers.
5 Likes
Searched through the files for a bit and couldn’t find it. I may have been wrong with it being Chika. If i come across it again, I’ll note where.
1 Like
It seems like going forward I’ll need to make distributions with both the original render and the inpainted renders. I really don’t want this to be such a divisive issue and this is the only solution I can come up with. A few people have already asked me about it, so the demand is clearly there.
5 Likes
Just want to make sure that the latest version on itch is the most update to date free version with new stuff correct?
At this point i think we all know AI can’t keep up the consistency and it always do something weird with hands lmao. Correct me if i’m wrong but if you’re checking all the details one by one i think you’re gonna waste more time checking if everything all right than making some progress with the game. I mean surely you did the same thing with the old renders and that’s why you’re taking this approach, tbh i don’t have any problem with using ai or making the images by yourself, good luck with that tho.
1 Like
Ok i found my answer. My next question is there a way or a guide that we can follow or anything i kinda dont want to miss anything or be locked out
I haven’t finished everything I want to in the update. But so far my partial impression about the art change is, the AI art is a negative for me. And it impacts my overall enjoyment of the experience. I think some of the renders looks extremely good and could be easily seen as an overall positive. But the consistency even within the same scene can vary greatly. Take Akane’s second dream sequence for example.
The problem with most images to me is that I feel you lose a lot of the emphasis on the characters actual weight in renders(as it can appear to fluctuate). If I had not played the game prior certain elements such as, characters losing weight or gaining it within some of the earlier stages aren’t as pronounced because of the variations within scenes. And I feel like this is an overall issue. Say what you want about the looks of the original style(which I love), daz, or other 3d models but consistency between renders, especially regarding weight is something that isn’t an issue. Akane sometimes within early stage scenes appears heavier then what I think is intended within certain scenes.
I think the looks of the characters faces with AI can be great but especially some of the one off scenes.But there are just dips in quality between different events. Or just outlines like the volleyball girls renders at school. Or the first time we see Chika stuffed by Erika. You lose details like folds or rolls that feel correct (in some scenes) stretch marks, and other nice details.
I hate to sound completely negative there are positives for the style change but the negatives out weigh that for me. And this went from a project that I was happy to support to one that I’m honestly uncertain about. Doing two different versions could be a solution, sure. But that seems like at least twice the work and defeats the AI arts original point. Either way regardless of my opinion I hope you figure out something that works for you and the supporters.
10 Likes
Personally I feel the AI art to be a huge downgrade. Characters used to be quite distinct and easy to recognize and now they look alike in a lot of scenes. There are also so much wonk; characters absorbing parts of eachother, characters changing appearence and weight across the same scene, hands and clothes looking off or clipping. Honestly I’d much prefer a slower release schedule with the original art style (I also find clipping and other issues were a lot less appearent in it).
Beyond the art I enjoyed the new content added, would love to get a changelog included with updates so I could be sure I actually got it all
8 Likes
If you’re going to use AI generated art moving forward, you should at least make sure what’s generated is consistent. Having played it now, the differences even just frame to frame in the same scene are jarring, and image quality is all over the place. A lot of the generated images look downright amateurish with visible chunks of white backgrounds, weirdly flat appearance, etc.
If what we’ve got now is indicative of what we’ll be receiving in the future then I’d rather you not artificially increase your workload. I understand the original 3D art has its own issues, but I personally found it distinct and refreshing compared to all of the other games on here that just reuse the same old DAZ assets. The jank is part of its charm, imo.
5 Likes
Yeah I’m aware fam. 
A good set of checkpoints, negative TIs, and the right Lora goes a long way. Unfortunately some things are still best handled via inpaint.
i’m going to be honest here after thinking about it… the ai art change while may of been a good idea to save you more time with design scenes however honestly with your style of design scenes for the story i honestly think you would be better off not continuing down this path while it might save you time the quality might suffer overall and would take away the unique style you originally made.
4 Likes
Hello! I’m very excited to try the new version. I will give my comments after playing. However, I must say,
I am strongly opposed to AI art, especially from someone as talented at you as making cute models. I desperately wish that the patreon was used by you to provide updates.
I wish you told us on patreon that the models were giving you problems, you’d been debating using AI art.
Hell, I would’ve argued that you cut back on content. Cut out school and tsumugi and the onsen, for example, and just focus on models you’re happy with for the basic main cast, and for a finite number of stage levels. Taking it slow and steady in less directions would’ve been preferable to this change.
However, I’m glad to hear you’re back, hope you’re well, thankful for being kept in the loop despite my patreon sub ending a month or two back due to college issues, and will provide more in-depth feedback after playing through.
10 Likes
My issue with ai generated images is the inconsistency they create. it’s jarring when hair colour, style, clothes, eye colours and number of limbs change in different scenes. Some images work great when they align with the old image layouts but the newer scenes have odd proportions, like one had 2 belly buttons, foot bending at shin. But in the images there is a style that blends with what you used to do with a smooth finish From the ai generated images. Maybe that Will improve but I’m sure people will agree they would prefer you to take longer with less content being high quality than shortcuts that cheapens the overall experience. Poll it on patreon, ask people who support you, communicate them, people are willing to accept less for a better experience
9 Likes