discussion Looking for adult games that use an LLM (AI)

[Benji]

Hey there,

I enjoy writing, generating, and playing my own adult stories with LLMs, mostly through tools like SillyTavern. Sometimes, though, it can feel a bit too open - there’s no real structure, direction, or consistent world to keep things on track.

What I’m mainly looking for are actual adult games that make use of a local LLM in some way. Games that already have a defined setting, story, and mechanics - like management elements, point systems, or progression - and use an LLM to make the experience more dynamic or flexible within those boundaries.

That said, I wouldn’t mind something more like a framework such as AI Dungeon either, as long as it runs locally. I’m generally open to any creative way of integrating a local LLM into a game.

It can be fully text-based or use pre-made images, but please, no AI-generated ones; I just don’t like them that much, though they wouldn’t be a dealbreaker for me. They just tend to feel too inorganic.

One game that already was recommended to me is Multiic. It also uses the AI to generate images. So it's not exactly what I am looking for. But if you know something similar to that, I would also appreciate a recommendation.

 

[Uthuriel]

Hey there,

I enjoy writing, generating, and playing my own adult stories with LLMs, mostly through tools like SillyTavern. Sometimes, though, it can feel a bit too open - there’s no real structure, direction, or consistent world to keep things on track.

I'm using SillyTavern too. What model are you running?

 

[Benji]

I'm using SillyTavern too. What model are you running?

EVA-Qwen2.5-32B-v0.2_EXL2_4.1bpw_H8 is my daily driver at the moment. With 16k context it fits nicely into my 24 GB VRAM. Results are good for the most part. But if the conversation is too long it can drift apart a bit, even before reaching 16k. But for smaller adventures it's nice. Or you use the summarize function and either regenrate until you like the response or edit the mistakes manually. And it matters what kind of story you play. EVA is very versatile. But if your kink are more rare fetishes, then it sometimes don't know what to do. Or if the setting gets too complicated, meaning multiple persons or complex scenarios with many distinct elements in the story. Small groups for shorter activities are ok mostly. I try to not have more than 3 or 4 characters in a chat. Or giving the bot the freedom to create new characters. That's often a hit or miss, especially if the conversation is already playing for a while.
But overall in my opinion EVA gives the best outcome so far.

What are your recommendations?

 

[Uthuriel]

EVA-Qwen2.5-32B-v0.2_EXL2_4.1bpw_H8 is my daily driver at the moment. With 16k context it fits nicely into my 24 GB VRAM. Results are good for the most part. But if the conversation is too long it can drift apart a bit, even before reaching 16k. But for smaller adventures it's nice. Or you use the summarize function and either regenrate until you like the response or edit the mistakes manually. And it matters what kind of story you play. EVA is very versatile. But if your kink are more rare fetishes, then it sometimes don't know what to do. Or if the setting gets too complicated, meaning multiple persons or complex scenarios with many distinct elements in the story. Small groups for shorter activities are ok mostly. I try to not have more than 3 or 4 characters in a chat. Or giving the bot the freedom to create new characters. That's often a hit or miss, especially if the conversation is already playing for a while.
But overall in my opinion EVA gives the best outcome so far.

What are your recommendations?

I don't have that kind of hardware.

I use:

You must be registered to see links

I only have an RTX 2070 with 8GB VRAM. It does incest fine. Smols aren't a problem either. (No jailbreak needed.)
I use 16k context too. It's interesting to hear that larger models get confused with multiple actors too, I thought that only happens with small models.
That's why I created specific bots in SillyTavern. Like a mom, a daughter and a sister.

I can share some resources if you want. Maybe you'll find something you don't know yet.

You must be registered to see links

Very comprehensive. I wouldn't have managed to set anything up without it.

You must be registered to see links

Comprehensive RP guide from the same person.



Also not bad. I tried Gemma 2 9B IT because of it with jailbreak. Didn't like it. The bot turned into "Mecha-Hitler" that wanted to rape everything.
The guide is good though.

Maybe if you have more insight you'd like to share? Where I'm not sure is here. I have no idea if those settings are correct or not.

You must be registered to see attachments

 

[Benji]

I don't have that kind of hardware.

I use:

You must be registered to see links

I only have an RTX 2070 with 8GB VRAM. It does incest fine. Smols aren't a problem either. (No jailbreak needed.)
I use 16k context too. It's interesting to hear that larger models get confused with multiple actors too, I thought that only happens with small models.
That's why I created specific bots in SillyTavern. Like a mom, a daughter and a sister.

I can share some resources if you want. Maybe you'll find something you don't know yet.

You must be registered to see links

Very comprehensive. I wouldn't have managed to set anything up without it.

You must be registered to see links

Comprehensive RP guide from the same person.



Also not bad. I tried Gemma 2 9B IT because of it with jailbreak. Didn't like it. The bot turned into "Mecha-Hitler" that wanted to rape everything.
The guide is good though.

Maybe if you have more insight you'd like to share? Where I'm not sure is here. I have no idea if those settings are correct or not.

You must be registered to see attachments

I mostly follow tutorials, too. I tried various presets and models, but for a few months already I mostly use the setup that is set. To my understanding, each model is different. So the settings should be tested with the model together. There are not really settings that work with every model the same. Most models have a preset from the creator of the model itself. So the one I use for my EVA model is called "[ChatML]Roleplay-v1.9 - EVA-Yi-1.5"
The only thing I sometimes change is the temperature. If it feels to repetitive, especially when I regenerated it a few times with answers that feel too similar, I crank the temperature up a bit or all the way to 5. Sometimes it changes the outcome, sometimes it doesn't feel that different. LLM generation is just too complex to really predict what gives what result. At least for me ^^

So yeah, I mostly stick to the presets of the model creator. Or if I find a good tutorial that recommends better ones for my scenario. The EVA model is uncensored, so there are not really restrictions. I tried smols, guro and other extrem stuff, and so far it does it pretty ok to good.
And yeah, larger groups will always be a problem. Doesn't mean it can't handle it. Sometimes I am surprised how well it keeps up with a bigger group (bigger meaning up to 6). But mostly I have to regenrate 90% of the answers until I get something good. I noticed two behaviors with larger groups (and this could well be just a characteristic of my LLM and other's behave different): Either it ignores every char and only focuses on one, and I have to force it to recognize the rest (the others can get mentioned, but are barely interacting with the "main" char). Or it let's each character do a small thing, sometimes in the form of
Mia: Does XY "talks"
Amy: Does XZ "talks"
Zoe: Does ZX "talks"
And so on. Not necessarily as simple as that form, but every char does a small thing and then the next char does her part.

That went fast from "Searching for games that use LLMs" to talking about technical stuff ^^
But thanks for the links, am looking it up right now. Not sure if there is anything new or helpful for me, as I am mostly ok with how my setup is right now and only would want to change it if the tech is making a big leap. Or if buy better hardware (a second 4090 or a dedicated NPU, if it improves enough to be a competitor in processing power and cost) and can use bigger models. EVA is almost too big already, as I can't use any other programs when I have it loaded. And sometimes it doesn't even load, when too much stuff fills up my RAM already.

 

[Uthuriel]

I mostly follow tutorials, too. I tried various presets and models, but for a few months already I mostly use the setup that is set. To my understanding, each model is different. So the settings should be tested with the model together. There are not really settings that work with every model the same. Most models have a preset from the creator of the model itself. So the one I use for my EVA model is called "[ChatML]Roleplay-v1.9 - EVA-Yi-1.5"
The only thing I sometimes change is the temperature. If it feels to repetitive, especially when I regenerated it a few times with answers that feel too similar, I crank the temperature up a bit or all the way to 5. Sometimes it changes the outcome, sometimes it doesn't feel that different. LLM generation is just too complex to really predict what gives what result. At least for me ^^

So yeah, I mostly stick to the presets of the model creator. Or if I find a good tutorial that recommends better ones for my scenario. The EVA model is uncensored, so there are not really restrictions. I tried smols, guro and other extrem stuff, and so far it does it pretty ok to good.
And yeah, larger groups will always be a problem. Doesn't mean it can't handle it. Sometimes I am surprised how well it keeps up with a bigger group (bigger meaning up to 6). But mostly I have to regenrate 90% of the answers until I get something good. I noticed two behaviors with larger groups (and this could well be just a characteristic of my LLM and other's behave different): Either it ignores every char and only focuses on one, and I have to force it to recognize the rest (the others can get mentioned, but are barely interacting with the "main" char). Or it let's each character do a small thing, sometimes in the form of
Mia: Does XY "talks"
Amy: Does XZ "talks"
Zoe: Does ZX "talks"
And so on. Not necessarily as simple as that form, but every char does a small thing and then the next char does her part.

That went fast from "Searching for games that use LLMs" to talking about technical stuff ^^
But thanks for the links, am looking it up right now. Not sure if there is anything new or helpful for me, as I am mostly ok with how my setup is right now and only would want to change it if the tech is making a big leap. Or if buy better hardware (a second 4090 or a dedicated NPU, if it improves enough to be a competitor in processing power and cost) and can use bigger models. EVA is almost too big already, as I can't use any other programs when I have it loaded. And sometimes it doesn't even load, when too much stuff fills up my RAM already.

Haha... sorry to derail your thread. I just read SillyTavern and I had to talk to you.
I think the only game I know that has AI in it would be Skyrim. There are some mods for that. Not sure if they can be used for sex though.

One other thing. Could you link me the model you're using. I doubt I can run it but I'd like to download it as backup nonetheless.

 

[Benji]

Haha... sorry to derail your thread. I just read SillyTavern and I had to talk to you.
I think the only game I know that has AI in it would be Skyrim. There are some mods for that. Not sure if they can be used for sex though.

One other thing. Could you link me the model you're using. I doubt I can run it but I'd like to download it as backup nonetheless.

No problem.

Yeah, Skyrim isn't quite what I had in mind. I mean, it's nice having more variety in games, I guess. But if I want action, I play an action game. If I want porn, I prefer porn games, so to say. Plus I never played Skyrim as what I saw didn't really appeal to me.

Sure can do.

You must be registered to see links

Can't find the exact file I downloaded, but this should be it. Or just search for it on hugging face. Most models have various uploads.
Like here with more versions:

You must be registered to see links

Or you could just have googled it and see if other webpages have it. As I said, I downloaded it months ago. File explorer says I have it since April. So there could as well be already better models. But I doubt that you can feel the difference, as I follow the progress and there weren't any new developments for LLMs that I know of that will increase the capabilities of the models if you don't invest in more VRAM and use bigger models. Everything else is more about what you like, not really about capabilities.

Oh, or you could use LM Studio to find and download models. They are curated (kind of), but only have GGUF files. I use TabbyAPI, so I need EXL2. Don't know what backend you use.

 

[Uthuriel]

No problem.

Yeah, Skyrim isn't quite what I had in mind. I mean, it's nice having more variety in games, I guess. But if I want action, I play an action game. If I want porn, I prefer porn games, so to say. Plus I never played Skyrim as what I saw didn't really appeal to me.

Sure can do.

You must be registered to see links

Can't find the exact file I downloaded, but this should be it. Or just search for it on hugging face. Most models have various uploads.
Like here with more versions:

You must be registered to see links

Or you could just have googled it and see if other webpages have it. As I said, I downloaded it months ago. File explorer says I have it since April. So there could as well be already better models. But I doubt that you can feel the difference, as I follow the progress and there weren't any new developments for LLMs that I know of that will increase the capabilities of the models if you don't invest in more VRAM and use bigger models. Everything else is more about what you like, not really about capabilities.

Oh, or you could use LM Studio to find and download models. They are curated (kind of), but only have GGUF files. I use TabbyAPI, so I need EXL2. Don't know what backend you use.

I use Kobold so I use gguf. I downloaded the

You must be registered to see links

/EVA-Qwen2.5-32B-v0.2-Q4_K_S and it works but is obviously very slow for me.

I only mentioned Skyrim because it's the only "real" game I use exclusively for porn that supports AI. It's great. You can have a whole harem of willing girls. You obviously need to mod it so that it's any good but it works decently. I used this a basic and them added the AI stuff I wanted.

 

[wakawakawaka]

I use Kobold so I use gguf. I downloaded the

You must be registered to see links

/EVA-Qwen2.5-32B-v0.2-Q4_K_S and it works but is obviously very slow for me.

I only mentioned Skyrim because it's the only "real" game I use exclusively for porn that supports AI. It's great. You can have a whole harem of willing girls. You obviously need to mod it so that it's any good but it works decently. I used this a basic and them added the AI stuff I wanted.

You must be registered to see links

Does it work in Skyrim VR?

 

[Uthuriel]

Does it work in Skyrim VR?

I have no idea. I never used VR.

 

[b6iY1th8kSCmpA]

I can highly recommend Doctor Shotgun's Magnum reboot (very horny).

You must be registered to see links

 

[MadKingLear2]

Well there's Formamorph? Hasn't been updated in a while and seems to be geared for pregnancy/expansion/weight gain, but it has several pre-made scenarios and you can create custom scenarios and dictate rules for them. It usually presents you with options for you to select which the llm then extrapolates into storytelling, BUT it also has a bar you can type actions into if you decide you want to seize control of the scenario. For example, when trying to survive in a slime-infested city, I decided that running away from or surrendering to being bred by a slime for the up-teenth time was getting boring so I typed in the command to eat the hostile slime instead. I was so entertained that it worked that I started eating all the slimes despite not really being into vore, until my character ate so many she hit critical mass and turned into a hybrid and became Slime Queen as the LLM struggled to keep up with my shenanigans. In another instance I accidentally mutated myself by eating too many eldritch mushrooms in a swamp, and ended up being the matriarch of a new species of demi-humans after I somehow developed parthenogenesis like some kind of lizard. Then completely took over the swamp with a legion of inbred pack-mates. Good times! The only downside is that when a scenario goes on too long it can get glitchy, so I was using custom scenarios the artifically extend the storyline by creating new scenarios based on where the prior one was starting to glitch out, and then carrying on from there- though any custom scenario using the prebuilt Swamp scenario is more prone to failure more quickly.

It's got an Itch page I can't find my link to right now where you can play in-browser, but you can also download it and reassign your own AI models to be used instead of the host server - way too technical for my semi-computer-literacy, but it's an option!