Easter Sale - 70% OFF: Create and FUCK Your Own AI Girlfriend! TRY FOR FREE
x
4.70 star(s) 10 Votes

Jake James

Newbie
Dec 12, 2021
16
7
Also, hate to bug you so much, but you might want to double check the reupload is actually using the same link and not a new one. Just tried redownloading, and it's the exact same file as I downloaded earlier today.
Hmm, weird I didn't know it worked like that. I edited the post now, it should have this link:
 
Jan 3, 2022
17
11
Snide_Krul & randomguyspain so I have been trying to find the max size AI my computer can handle which seems to be the Mistral-7B-Instruct-v0.3-abliterated.Q5_K_M. I only have a 8GB VRAM and anything offloaded outside of it makes the time spent generating responses unbearable. However, moving from a Q4_K_S to the Q5_K_M made a massive difference in regards of the AI's creativity and it doesn't seem to get stuck in those weird loops as much. Though my context is smaller by a bit so I can't go as deep into stories before its starts slowing down. But id recommend looking at the other ones you can run and see if there is Quant level higher you can run.
Thanks for the tip. I've got the same amount of VRAM (RTX 4060ti) and the (Mistral-Nemo-Instruct-2407-abliterated) Q5_K_M is definitely performing better than the Q4_K_S so far in my test run in Ashburn/Slime World. It ain't perfect, but it's definitely a significant improvement.
 

muoto8900

Member
Jul 7, 2018
127
63
Thanks for the tip. I've got the same amount of VRAM (RTX 4060ti) and the (Mistral-Nemo-Instruct-2407-abliterated) Q5_K_M is definitely performing better than the Q4_K_S so far in my test run in Ashburn/Slime World. It ain't perfect, but it's definitely a significant improvement.
Glad I could help. I tried the Nemo-Instruct-2407 as well but the performance was slow but the Q5_K_M is much faster for me though, it still has the issue of getting information stuck in it where it uses the same lines or paragraphs over and over. I think this happens when the context is getting too full. I have been trying to figure out a way to export a summary from the AI to see if I can mostly keep the story going but allowing me to reset my AI and clear its context of all the extra crap. I also need to test to see which context overflow method would be better as maybe rolling shutter will allow it to keep its most recent info but forget the beginning of the story.
 

Jake James

Newbie
Dec 12, 2021
16
7
Glad I could help. I tried the Nemo-Instruct-2407 as well but the performance was slow but the Q5_K_M is much faster for me though, it still has the issue of getting information stuck in it where it uses the same lines or paragraphs over and over. I think this happens when the context is getting too full. I have been trying to figure out a way to export a summary from the AI to see if I can mostly keep the story going but allowing me to reset my AI and clear its context of all the extra crap. I also need to test to see which context overflow method would be better as maybe rolling shutter will allow it to keep its most recent info but forget the beginning of the story.
I have another idea that I've been testing that may help. Simply remove the repeated info feed from the Choices Prompt and/or Stat Updates Prompt.

This is my Choices Prompt which you can tweak and play around with:
You don't have permission to view the spoiler content. Log in or register now.

I just use the default stat one since I'm not messing with those just yet, but you can remove the <INFO> text from there too

Sure, it won't know about characters or locations not mentioned before when outputting actions, but I find that can actually be a good thing since sometimes the AI would provide actions that have nothing to do with the text above it. This should help with reaching context limits.
 

Serpream

Newbie
Jun 7, 2018
69
22
So many different ways people are trying to rangle the AI back in line. I will edit and upload my rough version in few hours. Kinda just going in cricles now trying to refine and will see what people can do to help with.
 

Triforcedragon123

New Member
Sep 1, 2020
3
0
I have been trying to get LMStudio running with Nemo_Instruct (Q4_K_S) myself and while I can connect it to Formamorph successfully and get a few responses, I keep getting the following error when running "Client disconnected. Stopping generation... (If the model is busy processing the prompt, it will finish first.)" and I was wondering if any of you have encountered that and if so, what to do, if anything?

My setup has a NVIDIA GeForce RTX 3060 with 12gb VRAM.

Also unrelated to my current issues, the stuff being created here looks amazing.
 

Serpream

Newbie
Jun 7, 2018
69
22
So I have worked extremely hard on this. Between medical issues and lack of sleep it has been a welcomed distraction.
If people find ways to improve anything I have here then please provide it back to community. We are all learning and it will help everyone.

I would still say this is early Alpha, but starting to work great narrative wise. Each location is a type of flavour with same prompts. I wanted AI to create weird stuff, and it does sometimes now in a controlled, narrative sensible way.O-o

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

My link

*Reuploaded all of above with amendments that have made a huge difference and even had a good long story unfold.
Next step will be adding surivial elements back in.
 
Last edited:

TalRash4

Member
Feb 27, 2023
333
327
Note for self: adding player's clones into the slime outbreak world as NPCs was a fun mistake, that turned a somewhat grime survival into a tactical action rpg with a lot of fucking.
 
  • Haha
Reactions: nightsight
Sep 1, 2018
49
25
Snide_Krul & randomguyspain so I have been trying to find the max size AI my computer can handle which seems to be the Mistral-7B-Instruct-v0.3-abliterated.Q5_K_M. I only have a 8GB VRAM and anything offloaded outside of it makes the time spent generating responses unbearable. However, moving from a Q4_K_S to the Q5_K_M made a massive difference in regards of the AI's creativity and it doesn't seem to get stuck in those weird loops as much. Though my context is smaller by a bit so I can't go as deep into stories before its starts slowing down. But id recommend looking at the other ones you can run and see if there is Quant level higher you can run.
I tried a better version and it actually upgraded a lot. It has a lot of creativity, and it also seems to understand more the event you are trying to create, it does not deviate so much from the input you give.
It also almost collapsed my pc for not paying atention to the ram. I almost filled the ram capacity completely :LOL:
 

muoto8900

Member
Jul 7, 2018
127
63
I tried a better version and it actually upgraded a lot. It has a lot of creativity, and it also seems to understand more the event you are trying to create, it does not deviate so much from the input you give.
It also almost collapsed my pc for not paying atention to the ram. I almost filled the ram capacity completely :LOL:
Jeez, how long do your prompts take to generate now? I actually moved to a better one myself and even though LM Studio says it should partially offload, it seems to be all in my VRAM which is interesting. I am now using mistral-nemo-instruct-2407-abliterated@q4_k_s
 

Jake James

Newbie
Dec 12, 2021
16
7
So I have worked extremely hard on this. Between medical issues and lack of sleep it has been a welcomed distraction.
If people find ways to improve anything I have here then please provide it back to community. We are all learning and it will help everyone.

I would still say this is early Alpha, but starting to work great narrative wise. Each location is a type of flavour with same prompts. I wanted AI to create weird stuff, and it does sometimes now in a controlled, narrative sensible way.O-o

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

You don't have permission to view the spoiler content. Log in or register now.

My link World File

*Reuploaded all of above with amendments that have made a huge difference and even had a good long story unfold.
Next step will be adding surivial elements back in.
I'm curious as to what the Tables sections do. What's the output you get with those and what are their purpose?
 

Serpream

Newbie
Jun 7, 2018
69
22
I'm curious as to what the Tables sections do. What's the output you get with those and what are their purpose?
Rather than me explain, cheat type 'Display Tables' :p

Purpose is to track and keep silently reminding AI to check the data in tables to keep track of various stuff on player. Like Traits, what they do, how they effect player for events. Plus mutations, body part stuff and inventory for possible alternative solutions to events.

Best thing I could do without a local database to track the info.
 

fierylion

Member
Game Developer
Jan 18, 2019
223
419
Rather than me explain, cheat type 'Display Tables' :p

Purpose is to track and keep silently reminding AI to check the data in tables to keep track of various stuff on player. Like Traits, what they do, how they effect player for events. Plus mutations, body part stuff and inventory for possible alternative solutions to events.

Best thing I could do without a local database to track the info.
This is interesting, does it work consistently enough for you? I'm working on player notes system (the AI will look at and update player notes separately from the game text to avoid clogging up the game text every time), I plan to make the default notes format either table or some kind of list
 

Nouvi

Member
Jul 24, 2018
137
263
*one week later*... Having been pretty much plastered to Jake's Futanari Fantasy world since it was uploaded, and I changed to the mistral AI, I'm starting to wonder if that thing is being secretly developed by a group of the kinkiest fetishists on the planet, or being deliberately trained based on a fetish porn dictionary. ...and I'm kinda loving it, with how it makes every run end up with silly shenanigans. Definitely a new favorite toy to mess around with. (Also, screw the AI haters now, this is just too fun)

You don't have permission to view the spoiler content. Log in or register now.

Now you people might say "But you gotta have guided the AI hard for that stuff"... thing is, I didn't. While I did on the original worlds when I used qwen2, just to getting around the censorship into light fetishistic stuff, with mistral I've found it much more interesting to just use "continue", "wait", "next" or the like and let things play out, sometimes using the generated prompts and very occasionally, if prompts do not cover it, nudge it towards points of interests in the text like sounds, paths or landmarks.
I've also tried to look through Jake's worldbuilding/rules/prompt settings and baffled that there's basically no explanation for why the AI would go down such routes, but occasionally contradict the very clear, basic rules instead. Almost like Jake's "gief nsfw" world rule revealed the true kinks buried within that particular AI.

TL;DR: Crazy AI. Fun shit. Hope it continues like that.

Thank you fierylion, thank you Jake James and thank you all of you who end up making more fun worlds. Now I'm off to try out Serpream's world.
 

Jake James

Newbie
Dec 12, 2021
16
7
*one week later*... Having been pretty much plastered to Jake's Futanari Fantasy world since it was uploaded, and I changed to the mistral AI, I'm starting to wonder if that thing is being secretly developed by a group of the kinkiest fetishists on the planet, or being deliberately trained based on a fetish porn dictionary. ...and I'm kinda loving it, with how it makes every run end up with silly shenanigans. Definitely a new favorite toy to mess around with. (Also, screw the AI haters now, this is just too fun)

You don't have permission to view the spoiler content. Log in or register now.

Now you people might say "But you gotta have guided the AI hard for that stuff"... thing is, I didn't. While I did on the original worlds when I used qwen2, just to getting around the censorship into light fetishistic stuff, with mistral I've found it much more interesting to just use "continue", "wait", "next" or the like and let things play out, sometimes using the generated prompts and very occasionally, if prompts do not cover it, nudge it towards points of interests in the text like sounds, paths or landmarks.
I've also tried to look through Jake's worldbuilding/rules/prompt settings and baffled that there's basically no explanation for why the AI would go down such routes, but occasionally contradict the very clear, basic rules instead. Almost like Jake's "gief nsfw" world rule revealed the true kinks buried within that particular AI.

TL;DR: Crazy AI. Fun shit. Hope it continues like that.

Thank you fierylion, thank you Jake James and thank you all of you who end up making more fun worlds. Now I'm off to try out Serpream's world.
lol, I'm actually so glad it's been working well for someone else.

And yeah, Mistral is definitely a good one. It's very raunchy and if you like that verbiage it's perfect. Just depends on preference. I've been testing different model out recently instead of working on the world atm. Trying to figure out what model I like the most before continuing to wrangle it.

Funny thing is, I defined undercarriage use in the Dryad's blueprint so that's where it gets that for Amari, but if you start a chat with the AI in LM Studio it actually already kinda knows what it is. I just fleshed it out a bit.

As far as the flower? Idk, at one point I had multiple children pregnancies defined for Growli and maybe i still do, so it prompt bled those together which is perfectly fine. Add the fact that Freya is one of Ysera, the Life Bringer's (yeah, another WoW reference I'm so creative) chosen and it really likes to include nature and flowers with her.

That's all to say, I'm really glad to see it is indeed reading what I wrote and incorporating it in really interesting ideas for people other than just me and not reading everything verbatim.

With Sethi, have you encountered her special event? I feel like you'll know if you have. That's really the only other thing I'm curious to know is working for others. There's a few events in Teldorill as well, but idk if you like those characters.
 

Jake James

Newbie
Dec 12, 2021
16
7
Rather than me explain, cheat type 'Display Tables' :p

Purpose is to track and keep silently reminding AI to check the data in tables to keep track of various stuff on player. Like Traits, what they do, how they effect player for events. Plus mutations, body part stuff and inventory for possible alternative solutions to events.

Best thing I could do without a local database to track the info.
I'm in deep downloading and testing models, but I'll take a look at some point. I haven't messed with meta prompts/commands yet so that'll be a whole new thing to get frustrated with lol
 

Serpream

Newbie
Jun 7, 2018
69
22
This is interesting, does it work consistently enough for you? I'm working on player notes system (the AI will look at and update player notes separately from the game text to avoid clogging up the game text every time), I plan to make the default notes format either table or some kind of list
I keep refining it, but overall I gave a good 2 hour run yesterday and seemed to have kept track well.
I was hoping there was a way to offload the Traits back into the app, but settled for this for now.

The only thing I know that can break it is when player writes a fully custom prompt that ignores the current event, but a good 80%+ of time AI will recheck and rebalance the tables again.

My build was purely built to push and test AI with dynamic generated stuff. I will be working on the mechanics next week to see if I can get AI to really consider the flavour and intensity of the stuff in the tables, rather than the manual player checks required.
 

Nouvi

Member
Jul 24, 2018
137
263
As far as the flower? Idk, at one point I had multiple children pregnancies defined for Growli and maybe i still do, so it prompt bled those together which is perfectly fine. Add the fact that Freya is one of Ysera, the Life Bringer's (yeah, another WoW reference I'm so creative) chosen and it really likes to include nature and flowers with her.

With Sethi, have you encountered her special event? I feel like you'll know if you have. That's really the only other thing I'm curious to know is working for others. There's a few events in Teldorill as well, but idk if you like those characters.
The thing is, it was Freya, not a growli, and not just multiple children. It was clearly described as multiple pregnancy, in where every time she came, more and more impregnations happened, and children were added - like straight out of a fetish book. AI somehow even understood how that affected belly size and weight, and Freya's talk scenes like "Carrying five litters of my babies now".
My only explanation is that it actually saw multiple aspects, like her race and thinking of the saying "fucking like rabbits", all the quips you made for her, the Ysera part then tried to combine it, but slightly misunderstanding yet accidentally ending up looking like the real fetish. Freak chain of coincidences is all I can think of.

As for Sethi, yeah I've seen that event, twice actually, once with an elf and another time with a viera.
Though, of note, your personalities and how you've written things HAVE enabled some other fun shenanigans from the AI...
On two occasions the AI created it's own events for Sethi, one in which she invaded Freya's home and.... you can imagine.
Other time though was a bit more interesting... Sethi once again tried to pounce on Freya, just in her garden, but... tables got turned, and Freya ended up domming, breeding her full of babies.
 
  • Haha
Reactions: Jake James
Sep 1, 2018
49
25
Jeez, how long do your prompts take to generate now? I actually moved to a better one myself and even though LM Studio says it should partially offload, it seems to be all in my VRAM which is interesting. I am now using mistral-nemo-instruct-2407-abliterated@q4_k_s
It doesnt take long. I would say it writes like 15-20% slower. But the jump was big. I went from mistral-nemo-instruct-2407-abliterated@q4_k_s to mistral-nemo-instruct-2407-abliterated@q5_1
 

Nouvi

Member
Jul 24, 2018
137
263
Serpream Been testing your word for a bit, and gotta say I'm pretty impressed with how consistent you've managed to make it even if 'stress testing' it with vaguer commands than usual and relatively long sessions. Good work on that.

Just felt that my last experience leaves me with one question though... WHAT IN THE NINE HELLS DID YOU MAKE.
First thing I met was a friendly "small, orange furred, fox-like bat with expressive eyes" dropping down on my shoulder from above, and thought... "Nimbat much?"
Thing is, that a little later it put a seed into my character's stomach... raised my eyebrows a bit, but still a sorta interesting surprise to see the parasite mechanic happen in this way. Even if it felt kinda treacherous.
Anyway, seed hatched into a worm which isn't too special... worm evolved into a tiny serpent... interesting, but not all that special... later it evolved again into a WINGED serpent. That's where I start to feel it getting a bit off kilter.
Next stage though. That's the thing. It turned into a bloody dragon. Dragon, of all thing. So my character's now been running around with a miniature dragon squirming around in their stomach. That's beyond crazy. Craziest of it all, is that it's become more aware and actually communicating with my character. What have this all become...
 
4.70 star(s) 10 Votes