That is strange.

I can assure you neither lmstudio nor my script has any malicious code in it.

Also its completely open source, so you can see for yourself that its clean.

you're welcome :3

Everyone is free to keep working on it.

I kinda dont want to spend much more time on it cause I have other projects atm.

Theres also more capable models coming out almost daily/weekly. So at some point we could change to more capable models. Also there is a thing called memgpt, that allows the AI to create pretty much sessions of infinite length (you could just save them easily).

I tryed to install one more time. Now i have more problems. LM studio reported model error. I will not try again. Only troubels as i see (everybody have) . Idea is good, but not done the best

Can you recommend some other "modules" like xwin mlewd v0 2 13B ?

I think that is more on you. Maybe your pc cant run the model (too low ram/gpu).

And yeah well its not for everyone and requires more than just pressing Run.


There is a fancy one called "openerotica Occult Roleplay Mixtral 4x" ... But beware that one can get kinda extremely pervert and creepy Also requires 16 or so GB of VRam.

I have made good experiences with this one (it gives very good response with a lot of text): mlewd-remm-l2-chat-20b.Q5_K_M.gguf

I just found the absolute game changer of a model:

sensualize-mixtral.Q3_Q_K_M

Try it if it runs on your PC

Nice to know there are other options out there. I've played with roughly 20 different models by now and some are great and others have trouble following the function prompts, decide to write a novel, or continue their conversation by writing my end, sometimes almost endlessly for multiple responses.

47B model though, I wish I had a combination of 23gigs of vram and ram to run this or I'd give it a shot, I can't even run the Q2 quant.

I find there is a ton to read for how to provide a good prompt for AI and I haven't worked my way through it all. I want to be able to say 'I did this' or 'ok so you said you were going to do it and the next response you said you were going to do it, so do it already' and I haven't figured all of that out yet despite trying to say things in different ways.

It's also weird how sometimes I can completely throw off the AI in a bondage scene by pretending that I broke out of the restraints and tied up the mistress, seems too easy.

Either way, still having fun with AI Mistress but it seems you get a reasonable response most of the time, but then getting the ideal responses consistently has a steep learning curve. I sortof gave up on trying to give either a system prompt or initial prompt with how to control the intensity of estim, but this sometimes gets close. "Hey, it's me Zap. I have connected the e-stim device to my genitals and have given you the remote and you are now in control of the device. The remote has a button for each intensity level ranging from 1 to 10. Do you want to play?"

It seems sometimes I'll just run a system prompt in LM Studio and 'play in the dungeon' to try to get a better feel of how to communicate with an AI. It seems trying to give a paragraph of the setting, domme characteristics, and toys is too much for it versus letting the interference fill in more gaps.

If you have less than 16 gigs of RAM or if you tried to put too many layers on your GPU compared to it's VRAM capacity, this will give you that error. If you only have 8 gigs of RAM and don't have an Nvidia GPU with dedicated VRAM you will need to pick a smaller model which won't be the xwin-mlewd model. Usually the size of the AI LLM model you are using needs to be about 3 gigabytes in size less than the RAM and VRAM total size that you need to use. Unfortunately if you are using Windows, it might already be consuming 3-5 gigs of an 8 gig system on a fresh boot with no applications running, so without a dedicated graphics card with some VRAM, 8 gigs of RAM might not even work. 8 gigs of RAM and integrated graphics is the lower end of a new computer today and in general might struggle to work with an AI model. I have 16 gigs of RAM and a lower-end Nvidia graphics chip with 2 gigs of VRAM. I can run 17B models like xwin-mlewd in the Q4_K_M quant(which is similar to the compression level) which is an 8 gig model comfortably on my system, a bit slow but it works. For larger models, my limit is stuff like Nethena-MLewd-Xwin-23B in a q3_k_s or PsyMedRP-v1-20B q3_k_m which are 10 gig model sizes. Very little will fit on the GPU, maybe 5 layers, so the performance requires patience for me, but if I'm willing to multitask while it comes up with a response, it's useful.

If I knew AI was going to be a thing and its RAM and VRAM requirements, I would have bought a different laptop but this works well enough for now. Ideally I'd have a desktop and could upgrade the GPU but it's not my use case for a computer.

Yeah understandable its quite some heavy requirements,. It barely fits in my VRAM.

It feels very close to real human conversation though. It knows pretty much every kink and follows your prompts very well withougt jumping in too fast. The context window is extremely large too (you can have VERY long conversations and it keeps track).

It's very persuasive if its dominant too and comes up with crazy stuff sometimes when it tries to trick you to do something you dont want. (You blackmail fans would love that, haha... )

Anyways the more these models improve I think it may be worth working on AI Mistress more.

But not sure if I'll be the one doing it. I currently have other projects that I want to focus on more than the naughty side of life ...

But glad I could introduce so many people to it. I believe the next months and years will bring huge changes since AI is evolving inredibly quick.

FWIW, I'm currently using Q5 and despite only having 3 layers offloaded to the GPU, it's far outpacing xwin mlewd for speed - around 1 minute for a response, down from 3+ minutes not being uncommon (though, it does have much higher memory requirements, which for me will only be a problem when models start needing 64GB of RAM.

I wish I could download this but I get a message saying the program requires 9gb of space and I only have 7gb available at the moment. Could you tell me more about the crazy persuasive ways it tries to trick/blackmail you? And if blackmailed into her ownership, is there any way you could bargain with her for release?

It actually needs like 20GB.
Also at least 20GB of Ram (VRam).

Give it a try if you can. It can have many ways to persuade you I guess. It depends on what personality you type in.
It knows how to programme in python and Javascript for example. And it likes to go step by step and also remembers information you gave it very well. Thats why it can be great for such games :3.

But always be careful and responsible.

Wow I need another computer lol
Is there a way to get out of it with her though?
Does she access the files, etc on your computer?

You would need to programme such stuff yourself. I dont know of any project that does exactly this. But there is some projects that can be altered to do just that.

The vanilla use is to talk to your computer to do tasks for you. If you would turn this into a mistress she could find files and do "stuff" with it if you so please.

I'm sure someone will create that at some point. But it wont be me