/vrpg/ — I live in modded SkyrimVR

>>3964349 (OP)
This sounds absolutely amazing. Is there a way to get the voice mod working without a VR headset? I would definitely play this

>>3964397
Yeah my wife gets motion sickness so I did a flat screen version for her. Gate to sovngarde modpack was compatible, for sure. Super duper simple to just drag and drop. CHIM tutorial is very thorough, step by step. Give it a try!

>>3964349 (OP)
It's unbelievable what is possible today. RPGs have come a long way, but it's getting closer and closer to actually being transported into a another world.

Short ironman playthroughs is something I recommend everyone try, VR or no.

>muh visuals
Yawn. Wake me up when I can tell NPCs exactly what I want and get accurate in-character responses with narrative shaping effects.

>>3964532
Delphine tweaked when I told her I know about her sewer buddy and that she's in the blades. Started following me and I told her to fuck off but she wouldn't leave. Finally left when I told her the world would break if she didn't go home (probably)

Vampires in cages turned off hostility so I could rescue them, they would open their inventory to show that they didn't have a weapon to stab me in the back with, when I questioned if they were going to attack me at some point.

It's not all there, but it's 1000% ahead of any other experience available these days.

>>3964565
Holy shit that's pretty cool

>>3964615
I have no idea why I'm the only one talking about these emergent gameplay loops. I really wish more people knew. I feel like I'm going crazy.

>>3964734
I know now thanks to you. A while ago I saw a video about the AI dialogue, but I dismissed it, figuring it's still jank and superficial.
What you described sounds genuinely exciting.

>>3964565
oh wow, didn't expect an actual response. got any links?

>>3964349 (OP)
okay but does the AI react to rape from sex mods? Thats primarily what I use Skyrim for

>>3966232
Just search up CHIM-dwemer dynamics. You don't need a fancy modpack for it to work. For flatscreen special edition, the wabbajack pack recommend is 'gate to sovngarde.' It is quick, pretty, and good enough; For VR, mad god overhaul (MGO).

CHIM has a very simple thorough walkthrough.

>local LLM

Unless you're running two rigs with 5090s and using one of them for this you're not having a good time.

I'm waiting for Mad Gods Overhaul 4.0 then I'll dive back in. version 3.0 wasn't great

>>3966880
yeah, local LLMs are just not that powerful. is it 200b parameters? probably not. and that was what i consider a bare minimum, and it STILL got repetitive. 200b at least made it complex enough it could generate interesting, cohesive prose over many hundreds of responses.
anything capable of truly impressing me is going to have to run on a data farm, because i want trillion-parameter models.

I'll up your thread because i'm really interested. I never used chim. Local hosting a llm is out of the question because i have a mid machine with a 6700xt. How are the online llms? Would i be forced to pay for one, or the free ones are a viable alternative?

I never dabbled in AI so I really don't understand the free tier caps on groq AI. Will I be able to play normally or should I pay to use another one? If so, which of the supported ones would you suggest?

>>3964349 (OP)
Sounds cool, but you need like $7k hardware for this lmao. We're all poor af here.

>>3978058
You can always use online LLMs. Speaking of which, I wonder how much mileage can you actually get out of 5 dollars on claude api (1,6 milion input tokens, 300k output tokens). Like, how many hours would i get if i use it exclusively for chim before having to pay again

File: 0098819808ad9a703f6d9b704573f22c2b409a674936baadb3eb4988f03cfddf.gif (11 KB, 400x188)

>>3964734
Because the only people who talk about this stuff are isolated into niche communities online that only talk about this sort of stuff. Do you see a board that I can talk about the RPG/simulation/world I'm building that has over 100 individually crafted characters in the main cast, each with their own AI powered personality ect ect ect big world unique systems, every experience is different... can even play other TTRPGs inside the game by joining the LARP faction.
It's mostly text though. I'm still working on the GUI and images, I'm not using image-gen in the game, (I might add it to make portraits for spawned NPCs/mobs though) instead making custom scenes.... even tracking the character's schedules. Might even track if they need to poo, since I added around 50 restrooms.

Anyhow where do we post this shit?

File: 1689f0045d833450bfa02f63981d4bbcf6770298fdc60d1df56dd232fc8fcc4e.jpg (325 KB, 1920x1080)

>>3979838
That really depends on how much the LLM has to track. If it's just the conversation, then it's pretty cheap, probably a penny per response. You could use Haiku to really cut the cost down too if it's just the conversation.
Otherwise I'd say to try using deepseek 4 instead of claude, especially with 4.7 of sonnet/opus being trash that was only released to test their new mythos's guardrails for them.
>>3968705
local LLMs are as powerful as your machine can handle. You can always just pay for cloud GPU services though if your GPU is weak though.
Also, the issue isn't the power of the models you're using, it's what they're trained for. Try using a more specialized one. Though most of the RP models seem to want to rape the player.
>>3971377
>6700xt
You're fine, retard. You can run a models that are more than capable for this purpose, unless I'm misunderstanding the scope of what the LLM is doing for this mod.
You'll only run into issues if you're having a marathon length discussion with multiple characters, and that can be somewhat fixed with better chatlogging procedures.

File: 1754962479525751.jpg (76 KB, 900x900)

>>3964349 (OP)
I don't believe you. VR sucks to have on for more than an hour and a half tops. And discussions with LLM's are only entertaining for a few minutes. Sure, you might've had your "WOW THIS IS AMAZING" moment for a few hours, but this is not something anyone would enjoy doing for a long period of time.

>>3964349 (OP)
Can you give orders to followers? I've been enjoying having Teldryn as my bro, and this VR snd LLM thing sounds like next level shit to try in the future.

File: file.png (1.94 MB, 1200x1200)

How can we stop this? When will it end? how many indians has he hired to make these threads?

>>3964349 (OP)
How does the LLM mod work? Is there a guide for installing it? Is it on nexus?

>>3982404
>6700xt
>You're fine, retard. You can run a models that are more than capable for this purpose, unless I'm misunderstanding the scope of what the LLM is doing for this mod.
Do you seriously think that? Because i keep reading that you need a 4090/5090 gpu for the local LLM and using AMD instead of nvidia means going automatically 50% slower. It's hard to find someone who benchmarked this with a mid level graphics card. Fuck, you're going to make me reinstall the game and try it. I might even buy it after 3 or 4 thousand hours

Bump, I'd like to know other people's mileage with setting this shit up before i spend the next 4 days trying to get it to work

>>3964349 (OP)
post modlist?

>>3982404
>Though most of the RP models seem to want to rape the player.
You already sold me on it, you didn't have to sweeten the deal.

File: f285d2629a5b7448f6edb6a2d7bae515.jpg (801 KB, 3840x2278)

>>3982918
Look up Mantella, CHIM or SkyrimNet. First 2 are on Nexusmods, Net is on github.

>>3984580
https://www.canirun.ai/
Mixture of Experts (Moe) models have a great quality to speed ratio.

>>3990367
>MGO SkyrimVR (one click install) with three simple additional mods. Namely, CHIM, climb anything, and deadly damage.

>>3990477
Can you elaborate on MOE vs others? Which parameters should I prioritise choosing a model? I'm genuinely considering taking a week off to set this thing up offline

>>3990477
>Look up Mantella, CHIM or SkyrimNet.
Also, which one of the three would you suggest to use? I'm good at modding but i'm a complete noob in terms of AI shenanigans

>>3993230
>Can you elaborate on MOE vs others?
I don't really know how it works, I just know it runs 10x faster than a similarly sized "dense" model, without using as much vram. You should check out >>>/g/lmg for more info on local models.

I have a 4090 and was using "Gemma 4 26B A4B it"(one of the uncensored variants) in LM Studio. And was able to prompt it while also playing Marvel Rivals at max setting, and running video in the background, and only experienced negligible fps loss.
you do want to cut down the amount of gpu cores it offloads to by half though.
Slowed to an slideshow while using the 31B dense model.

I also tried the 4B version but the quality was way worse and it didn't run any better than 26B.
The 26B was able to give a great and accurate in-character info about gods in Elder Scrolls, while the 4B one was 50% bullshit.

I also tried other models that were supposed to be good at RP, but they all had trouble with hallucination. Gemma 26B and 31B were the only ones that didn't just make stuff up. Remember to search "abliterated" or "uncensored" if you wanna do lewd stuff.

>>3993232
I haven't actually tested them yet, since I'm still waiting for the Steam Frame to put everything together.

https://www.youtube.com/watch?v=g-hJh_Jiolo
https://www.youtube.com/watch?v=H_poAl8Zllk
This guy has a video comparing Mantella and CHIM. And then CHIM and SkyrimNet
To me SkyrimNet looks the most promising.

>>3993942
>SkyrimNet
why not just call it SkyNet

https://www.youtube.com/@BrainFrogVR/videos

How much of this guy videos are staged? I mean, I get the enthusiasm for these AI systems but there's no way shit like this can happen spontaneously

>>3994837
Everything.

>>3993930
>>3993942
you are convincing me to play this... I don't have a vr headset yet but if this works on my rig I'd definitely get on in the future. is there a certain version of skyrim I should pick up? and im using a 5070ti and a ryzen 7 5800x with 64 gb vram is that enough?

>>3994870
"Special Edition" is the latest one that all the mods are made for now. VR edition is just special edition with vr on top.
Steam summer sale is right around the corner so I'd get it then. GoG has it without the DRM but they don't have the VR one.

You have 64gb of RAM not VRAM. VRAM is on your gpu, and can run AI stuff faster than regular ram. You can use both though.
The 5070 ti should have 16 gb of vram. I honestly don't know how well it's going to run, you could download LM Studio and test it, or ask in /lmg/ on g. If it doesn't run well, then you could go with a smaller model. Or use OpenRouter, which I think is the default for these mods, but that's going to cost you, and you might have to jailbreak to circumvent censorship. The prices aren't bad though, and censorship shouldn't be an issue if you're not gooning, probably.

>>3995112
>You have 64gb of RAM not VRAM
Speak for yourself, peasant.

>>3995112
shit i meant ram, the v snuck in there, it's only 16gb of vram.
what gpu are you using? a 5090?

>>3995250
>what gpu are you using?
4090

>>3995462
gyaddamn... okay i doubt it'll work on my lowly 5070ti then. rip

>>3997237
I have a much shittier gpu and supposedly I can run some systems locally, according to https://www.canirun.ai/
See what results it gives to you. If I'm not mistaken you can also run part of the AI locally and demand the more computational-heavy parts to an online provider

I still have to try this though, I haven't had the time to set it up properly

So it's been a couple of months since OP posted. Did someone else try to set this up with a more forgiving setup? I really don't have the time to mess with another skyrim installation rn