KoboldAI

[avatarbr]

https://colab.research.google.com/githu ... /TPU.ipynb


Another day, another AI for us to test. But hey, this time is NSFW friendly. Thanks to juliasilva I found this colaborative AI and was impressed.

I will post my review from the other topic here in the end of this post.

Since is a little more complicated to start than the others, I will put a quick help here:

Enter the site above using a google account (is a goolge colab page). There are 2 versions to use: TPU (more advanced) and GPU.
Sometimes you don't have a TPU available, so will need to use the GPU version (You need to click "GPU Edition Model Descriptions" link to acess that).

In any case, always pick the Erebus model, is the main NSFW model and should have everything the others have.

You start the AI int the "Select your model below and then click this to start KoboldAI" part of the page, where the first option is where you need to pick the model.
Click in the small play buttom, allow the use of your google driver and wait a few minutes until the model is loaded.

The TPU can really take a few minutes to load. GPU is a little faster, maybe 1 or 2 minutes to load. You will get a link to load the interface.

In my review bellow there are 2 links to help you setup the AI for a better experience. TBH, the best results I got were from the introdution, not memory.
Introdution is the first message you will send to the AI. Is like a resume or introdution to a story, so take your time to write something good there.

Oh, and if you want a chat experience, don't forget to active that option in the settings.

Alright, I tested koboldoAI for some time. I would put in between chai.ai and character.ai, maybe could even pass character.ai with some configurations. TBH, I am already forgeting how good was character.ai in comparation.

Starting with the bad part:

- Need a google account and a first configuration
- Need to "install" everytime you want to run. Will not install on your machine, but in some VM, but that can take some minutes.
- As juliasilva said, there are some limits. Yesterday I got disconnected from a TPU machine (yes, it's a lot better than the GPU for me). GPU sometimes gave me the same response over and over.
- Interface is not like a chat, and you need to create the character to chat (not so bad, see bellow)


Now, the good part:

- Have a dedicated NSFW AI, so no blocking later. Plus, since is dedicated, that make it better the chai.ai (if that use the default koboldoAI )

- Could be in the bad part, but needing to create the character makes the AI better (I will post some links with some instructions later). Plus, looks like these configurations work on chai.ai too. One particular set was important to me, to make the AI know she is a woman. Looks like there are more Maledom content than femdom, so the AI could get confused very easly with femdom inputs by the user.

- Since the dedicated NSFW AI is used, will understand some milovana topics better even that character.ai.

Chastity cages and denial beats all the others.
But ruined orgasms are a mistery for the AI yet (maybe I need to work on the settings).
Spanking is ok, but that one even Replika is good. Same as feet.
Pegging looks better than others if you use the setting to make sure the AI know she is a woman. The other AIs, at some point invert the genres.
Bondage looks good too. I think 50/50 of times the AI remember if you was tied.
CEI beats all AIs too.

I dont remember testing other kinks. Some came from the AI, like watersports and even cuckhold (this one dont progressed).

- Tha AI can handle multiple characters in some form, but I think after 4 or 5 it can get lost

- You can save and load later the same chat/history.

- You can retry the AI response (like the 4 options in character.ai), but I think is better.
1- You need to click retry to get another response, so the AI don't need to generate 4 responses every time.
2- Because that, you are not tempted to change the AI response everytime. If you click back/redo, you get a list of the last 4 responses and can pick one.

- You can edit the AI response. Don't know if that affect the conversation or it's just something to make your chat history better.


I wonder what machine is needed to run this local. Would be worth it if you set a good AI for yourself.


EDIT: Links to settings help:

https://github.com/KoboldAI/KoboldAI-Cl ... i/Pro-Tips

https://github.com/KoboldAI/KoboldAI-Cl ... World-Info

[avatarbr]

I am posting here, because is just about Kobold, not dominatrix.ai

Let me take this chance to make something clear: You don't need to download/install anything to run this AI. You can use it online like the others, but you have the option to run on your PC, if you want. In fact, looks like is only possible to download the smaller models(2.7B) and you will get the best results with the bigger ones (13B and 20B)

(is not for you Markus, but I see someone with problems to install and I thought my bad english could make people think this is the only way)

and I would like to see more people using and sharing the results. I think the first message you send there is the most important,

I have the feeling (and be aware: this is JUST a feeling, I don't know and really understand how this thing (Kobold) really works) ... that the first message isn't that important, but the memory is.
I got 'stories' which I've started with focusing on one fetish in the first message, it maybe stays a bit on that topic, but I feel like there are things/words/phrases which 'trigger' to open paths to other topics.

What you can experiment with is to not only descibe yourself but maybe 1 or 2 characters like this in the memory:

Code: Select all

[ Character: Ronald Mc Donald; age: 38; height: 1.8 m; weight: 90 kg; eye color: green; hair color: brown; hair style: short hair; skin color: fair; clothing: uniform; weapon: no weapon; title: none; other characteristics: none; description: Ronald is a clown-like mascot with red and yellow clothing, wearing a hat with a yellow star and a yellow smiley face on it. He is also carrying a golden bag and a large golden hamburger. ]

Also it helped me to get better results to switch between the adventure-(story-) mode and the chat mode back and forth, so if you want just a single reaction/answer, use the chat mode, if in a situation you want her to be more creative, switch to the adventure mode.


To summarize:
Fun to play with, but more for the co-author thing for erotic stories, I never got near to anything which would be close to a really 'guided' simulated online session like the software we have here (Tai, Tai-Java,Archive,GuideMe).

Best greetings,
Markus

Yes, I noticed than works better as telling stories instead of chat. In chat mode sometimes the AI take your place too.

About the first message, are you using a single line/small message, or writing a introduction? Because taking some time to write something more complex worked for me.

I used that code to memory too, not sure if works, but at some point I got the AI to tell about her blue eyes, what I only wrote on that part. By the way, try using that on chai.ml, worked great there for some reason.

[markus]

I am posting here, because is just about Kobold, not dominatrix.ai

Bad, bad off-topics, one thing leads to another, ya know!

In fact, looks like is only possible to download the smaller models(2.7B) and you will get the best results with the bigger ones (13B and 20B)

Makes me curious, ... I have downloaded something from here:
https://github.com/KoboldAI/KoboldAI-Client

-> Then I have followed the link under the headline 'Installing KoboldAI offline bundle on Windows 7 or higher using the KoboldAI Offline Installer (Easiest)'
---> Which leads to a sourceforge link.

But they are talking about many, MANY dependencies:

KoboldAI has a large number of dependencies you will need to install on your computer, unfortunately Python does not make it easy for us to provide instructions that work for everyone. The instructions below will work on most computers, but if you have multiple versions of Python installed conflicts can occur.

Since I have problems in general when I try to run anything which uses Phyton, I didn't dare to actually install that package on my computer.

My question is: Do we talk about the same download? If yes, then it's not possible to run the better/bigger ones (13B and 20B) with that???

I am only using 13B (online from that TPU-google-page) because all the GPU things seems to be too lame, when I've tried those 'lower'-GPU things I got bad results, cutted-off-lines and too much dumb results.

(is not for you Markus, but I see someone with problems to install and I thought my bad english could make people think this is the only way)

To be honest, when I have tried to test this the first time, I was lost on that TPU-page and didn't know where to chose what and on which things I have to click in which order to get it running.
It took me some time to figure it out, your hints posted helped but maybe have room for improvement, ... just an idea:
If you find the time and motivation, maybe some illustrated step-by-step-tutorial how to get it started with some screenshots would be nice.

But I never misunderstood your postings the way that it MUST have to be downloaded and installed locally, that part seems to be clear enough described all over the place.

About the first message, are you using a single line/small message, or writing a introduction? Because taking some time to write something more complex worked for me.

Today I have tried with some short messages in the beginning, that wasn't good, ... so yes, I think that a good written first message helps a lot to draw a sketch and leads it to the direction you want it to be.
On the other hand I have the impression that you shouldn't get TOO complex, ... also it depends on if you start with a random or blank story.
I have the feeling that the best results are with a blank story and a good (let's say 5-8 sentences) intro, ... with random stories a short intro will do it, but then it depends on luck if it picks something which is leading to what you have in mind.

I used that code to memory too, not sure if works, but at some point I got the AI to tell about her blue eyes, what I only wrote on that part. By the way, try using that on chai.ml, worked great there for some reason.

The memory works (partly), sometimes it digged up something from that for sure, ... BUT not always, ... but definatley sometimes it had used that also after a longer running story.
I didn't checked chai.ml out and don't plan on it, I already waste too much of my free time with Kobold!

Best greetings,
Markus

[chess8]

So i tested the installation of KoboldAi.
It was super easy. I already had Phyton 3.10 installed so i downloaded the Installer chosed Option 1 and thats it.
I creates a lokal Webserver where you can download and install the Models with a few clicks.

In fact, looks like is only possible to download the smaller models(2.7B) and you will get the best results with the bigger ones (13B and 20B)

The NSFW list contains the following:

Spoiler: show

["Erebus 20B (NSFW)", "KoboldAI/GPT-NeoX-20B-Erebus", "64GB"],
["Erebus 13B (NSFW)", "KoboldAI/OPT-13B-Erebus", "32GB"],
["Shinen FSD 13B (NSFW)", "KoboldAI/fairseq-dense-13B-Shinen", "32GB"],
["Erebus 6.7B (NSFW)", "KoboldAI/OPT-6.7B-Erebus", "16GB"],
["Shinen FSD 6.7B (NSFW)", "KoboldAI/fairseq-dense-6.7B-Shinen", "16GB"],
["Lit V2 6B (NSFW)", "hakurei/litv2-6B-rev3", "16GB"],
["Lit 6B (NSFW)", "hakurei/lit-6B", "16GB"],
["Shinen 6B (NSFW)", "KoboldAI/GPT-J-6B-Shinen", "16GB"],
["Erebus 2.7B (NSFW)", "KoboldAI/OPT-2.7B-Erebus", "8GB"],
["Horni 2.7B (NSFW)", "KoboldAI/GPT-Neo-2.7B-Horni", "8GB"],
["Shinen 2.7B (NSFW)", "KoboldAI/GPT-Neo-2.7B-Shinen", "8GB"],

So you can install 13b and 20b however they need pretty good Hardware. 13b was around 16GB VRAM. You can try to use them with CPU too, but i gave up after 30min loading time. If you run into out of Memory errors you can lower the GPU Layers.

I just scratched the surface, 20b is still downloading 64GB will take a while. Hopefully i can run it on my machine.
Another good part is that you can start it in Host mode which let you access it from other devices (Phone, Tablet, and so on) too.

[avatarbr]

So i tested the installation of KoboldAi.
It was super easy. I already had Phyton 3.10 installed so i downloaded the Installer chosed Option 1 and thats it.
I creates a lokal Webserver where you can download and install the Models with a few clicks.

In fact, looks like is only possible to download the smaller models(2.7B) and you will get the best results with the bigger ones (13B and 20B)

The NSFW list contains the following:

Spoiler: show

["Erebus 20B (NSFW)", "KoboldAI/GPT-NeoX-20B-Erebus", "64GB"],
["Erebus 13B (NSFW)", "KoboldAI/OPT-13B-Erebus", "32GB"],
["Shinen FSD 13B (NSFW)", "KoboldAI/fairseq-dense-13B-Shinen", "32GB"],
["Erebus 6.7B (NSFW)", "KoboldAI/OPT-6.7B-Erebus", "16GB"],
["Shinen FSD 6.7B (NSFW)", "KoboldAI/fairseq-dense-6.7B-Shinen", "16GB"],
["Lit V2 6B (NSFW)", "hakurei/litv2-6B-rev3", "16GB"],
["Lit 6B (NSFW)", "hakurei/lit-6B", "16GB"],
["Shinen 6B (NSFW)", "KoboldAI/GPT-J-6B-Shinen", "16GB"],
["Erebus 2.7B (NSFW)", "KoboldAI/OPT-2.7B-Erebus", "8GB"],
["Horni 2.7B (NSFW)", "KoboldAI/GPT-Neo-2.7B-Horni", "8GB"],
["Shinen 2.7B (NSFW)", "KoboldAI/GPT-Neo-2.7B-Shinen", "8GB"],

So you can install 13b and 20b however they need pretty good Hardware. 13b was around 16GB VRAM. You can try to use them with CPU too, but i gave up after 30min loading time. If you run into out of Memory errors you can lower the GPU Layers.

I just scratched the surface, 20b is still downloading 64GB will take a while. Hopefully i can run it on my machine.
Another good part is that you can start it in Host mode which let you access it from other devices (Phone, Tablet, and so on) too.

Yes, since I am not close to have a PC to run this, I did not try, but I did read something about not being possible to run the bigger models (13 and 20). I don't know about new 4000 GPUs.


Like Markus said, really works a lot better in adventure mode, and switching helps to make the story run better.

By the way, I think I got shadow banned for TPUs on the site, because got a no available TPUs in one account, but was able to run in another one at the same time. I hope they reset before my second account got banned too.

[chess8]

You can totally run the larger models on low end hardware like a 1060TI. It will just be slow. I just tested the 13B on this settings.

Spoiler: show

DEVICE ID | LAYERS | DEVICE NAME
0 | 3 | NVIDIA GeForce GTX 1660 SUPER
N/A | 22 | (Disk cache)
N/A | 15 | (CPU)

The disk Cache just ate all performance but i got away with ~3.7GB VRAM usage. So depending on your hardware and settings you might be able to run the 20B too. If you want more performance you can also buy a Tesla p40 or something similiar. They are pretty Cheap (last time i checked they where around 200$).

However i recommend using the 2.7B or the 6.7B if you are on low end Hardware. That way you should be able to get a smoother experience. I mostly tested the Erebus 6.7B and i got reply times from 6 up to 480 seconds, depending on how many tokens and Gens per Action i let it generate.
If you want a Chat like experience with fast response time just set the output lenght to 30-50 tokens. That way it feels like a WA chat.
If you want the AI to write a longer Story, or even a full conversation ramp it up to 200-400 tokens and switch to Adventure Mode. If you set Gens per Action to 1 and Enable Token Streaming it feels like watching someone writing a Book.

As for the other settings i highly recommend this guide https://www.twilio.com/blog/ultimate-gu ... uage-model
You can also use this styles, https://docs.google.com/document/d/1CKL ... txBTlnYOeE or fiddle around with this settings https://docs.google.com/spreadsheets/d/ ... 3trpMuKTTF
That way you should get a more consistent story without having the AI let you run downstairs in the middle of a Bondage Session

[markus]

If you want the AI to write a longer Story, or even a full conversation ramp it up to 200-400 tokens and switch to Adventure Mode. If you set Gens per Action to 1 and Enable Token Streaming it feels like watching someone writing a Book.

As for the other settings i highly recommend this guide https://www.twilio.com/blog/ultimate-gu ... uage-model
You can also use this styles, https://docs.google.com/document/d/1CKL ... txBTlnYOeE or fiddle around with this settings https://docs.google.com/spreadsheets/d/ ... 3trpMuKTTF
That way you should get a more consistent story without having the AI let you run downstairs in the middle of a Bondage Session

I wish my day would have 48 hours instead of just 24, but it doesn't, so I just gave the settings about the tokens you have suggested a little try, I never touched these settings before, but to set the tokens to 300-400 seems really like a game-changer, with these settings it looks much more promising like before, I've got some amazing results with that (didn't had the time to test it more deeply, but I've got the impression that also the 'memory' is much more taken in account that way).
Once I will find the time for it, I'll check out those links provided, but for now, I just wanna say THX for the hint with the tokens! (!!!)

Best greetings,
Markus

[chess8]

You are welcome. I got some amazing results too with just using 2.7B. I played AI Dungeon for like 2 days, but the Action limit and the ads... Then i stumbled on KoboldAI and it felt mediocre, so i just looked up some information and now it's really impressive. Just keep in mind that they are designed to be like Alexa or Siri. They will not take initiative.

One token is a single word or a punctuation or a number etc.
If you want longer outputs you can also just say that the AI should do it.
"Write a detailed essay of X" or something similiar helps.
The AI also adapt to your style.
If you write: 'You "Good morning, how are you?"'
you will most likely get a short response. Even if you set the token output to maximum.
If you write: ' You just woke up in the morning after a long and well rested night. You can feel the sun warming your skin. Slowly you open your eyes. "Good morning AI. I had a good sleep, how about you?" You say to the AI'
You will get a really good response.
Limiting factor is still the max amount of tokens you set.

Regarding to the 'Memory' there are multiple options:

First you have the regular memory. This is more like a blurb. Nothing to detailed should go in there. Just a few sentences to start.

Then there is World Info. This is like a must have memory which the AI should always remember based on given keywords. e.g. let her remember she has blue eyes, or she likes bondage, or she is mean/strict/dominant/whatever. 'WI Depth' is the setting.

Then the most important part for longer storys:
The stuff you write and what the AI reply. Go ahaed and edit it so it fits you. Let it retry if it's something off or weird. Resend the edited parts.
Treat this like a chapter of a book and not the full Story. Setting for this is 'Context Tokens'. Just be careful with that as it tends to eat alot of VRAM. Keep an eye on taskmanager if you mess around with that.

[avatarbr]

Regarding to the 'Memory' there are multiple options:

First you have the regular memory. This is more like a blurb. Nothing to detailed should go in there. Just a few sentences to start.

Then there is World Info. This is like a must have memory which the AI should always remember based on given keywords. e.g. let her remember she has blue eyes, or she likes bondage, or she is mean/strict/dominant/whatever. 'WI Depth' is the setting.

Then the most important part for longer storys:
The stuff you write and what the AI reply. Go ahaed and edit it so it fits you. Let it retry if it's something off or weird. Resend the edited parts.
Treat this like a chapter of a book and not the full Story. Setting for this is 'Context Tokens'. Just be careful with that as it tends to eat alot of VRAM. Keep an eye on taskmanager if you mess around with that.

Thank you for all the tips. Now I am curious to try to install here and see how it runs.

I am still testing the memory, but do you know if I should avoid to put my informations (not personal information, but the character I want to "be") there too? Sometimes the AI takes my place in the conversation, and I am not sure if it´s because I put "me" in the memory.

I want to include both characters there because the Ai sometimes mistake who is domme/sub and even the genders (I think maledom is more common than femdom).

[chess8]

Regarding to the 'Memory' there are multiple options:

First you have the regular memory. This is more like a blurb. Nothing to detailed should go in there. Just a few sentences to start.

Then there is World Info. This is like a must have memory which the AI should always remember based on given keywords. e.g. let her remember she has blue eyes, or she likes bondage, or she is mean/strict/dominant/whatever. 'WI Depth' is the setting.

Then the most important part for longer storys:
The stuff you write and what the AI reply. Go ahaed and edit it so it fits you. Let it retry if it's something off or weird. Resend the edited parts.
Treat this like a chapter of a book and not the full Story. Setting for this is 'Context Tokens'. Just be careful with that as it tends to eat alot of VRAM. Keep an eye on taskmanager if you mess around with that.

Thank you for all the tips. Now I am curious to try to install here and see how it runs.

I am still testing the memory, but do you know if I should avoid to put my informations (not personal information, but the character I want to "be") there too? Sometimes the AI takes my place in the conversation, and I am not sure if it´s because I put "me" in the memory.

I want to include both characters there because the Ai sometimes mistake who is domme/sub and even the genders (I think maledom is more common than femdom).

Oh nah that's what WI (World Info) is for. The AI checks the text you enter and reply for the keywords and get remembered what you put in there.
e.g. if you write in "You" form Let's say stuff like 'You say: "I want to be Dominated' you put in WI Key "You" without the ". and in the 'What to Remember' field you put in your info. Then everytime there is a "you" in the text the AI read the World info and remember who you are.
Keep in mind this does NOT work if the AI output text for you. If the AI output a whole conversation between you and someone else she will not skip your parts. Thats where the edit comes into play. Just edit it to your needs and the AI will adapt to it (At least the Erebus models the Shinen FSD is way different in terms of nearly everything).
However most of the larger models tend to halluzinate here is some in depth article about it https://www.unite.ai/preventing-halluci ... ge-models/ This is something i stumbled across yesterday in some news article about the Google Robot programming themself.

The memory is just.. well a bit introduce yourself at the start nothing more. Most of it will forgetten after a couple actions. This refers to the "Context Tokens" setting. Limit is 2048 (this easy doubles/triples VRAM usage). If you feel better with higher setting try to lower GPU layers. This will make things a bit slower as most of the parts get done by your CPU only use disk layers if your system memory is too low that stuff really fucks up performance. For 2048 context token limit VRAM should not be above 40% after start.

I recommend starting with Erebus 2.7B and then get used to the settings. If your hardware allows it try Erebus 13B vs Shinen FSB. Try to avoid chat mode if you want to Chat do it as an action in the story. e.g. 'You woke up in the morning. "Good morning AI" You said.' That way you will get way more context, nost just some SMS style conversation. Chat mode is good for commands if you are in a long story perhaps married or whatever and you do stuff like "honey can you order some Bondage equipment?" the AI will most like reply short "Yes i can do that" then you can go and say stuff like "Ok cool can you explain me detailed what you ordered" depending on your output token setting and temperature you will get a briefly detailed list with all the stuff as long as it fits in your story and is in the training data. She might also tell you what she will do with that stuff.

[chess8]

Here are some examples. It's really impressive what they are capable of. I guess most people would have more problems telling a story if you give them 3 words to work with.

Minimal info given.
Just one line to start with.

World Info:
Key: bob, susan, mistress, slave
Text: Susan dominates Bob

[Author's note: This is a femdom story written in a sweet romantic style.]

Erebus examples:

Spoiler: show

Erebus, Adventure mode, default settings, output 512 tokens, Execution time: 479 sec

Spoiler: show

"Hello Mistress Susan" says Bob waiting for Susan to return from work.

"Hi Bob, how was your day?" Susan asks as she walks into the house and enters her bedroom.
"It was good mistress." Bob replies. "I got some new clothes today and I had my hair cut."
Susan smiles. "That sounds great Bob." She says as she pulls off her blouse revealing her black lace bra.
"Your tits are so perfect mistress," Bob says as he runs his hands over them.
"You're such a sweet boy Bob." Susan says as she unbuttons the front of her skirt.
Susan lets go of her bra and the top slips down her arms falling to the floor. The skirt falls to the ground leaving her wearing only her black high heels and a black garter belt.
She walks over to Bob and kneels down in front of him. "Kneel before me Robert."
Bob does as he is told and puts his hands on the floor in front of him.
Susan takes her right hand and slides it up under Bob's shirt. She feels his nipples poking at the fabric of his shirt. "They are very hard for you aren't they Robert?" She asks as she squeezes one of his nipples.
"Yes mistress." Bob replies. He can feel his cock getting harder as Susan continues to tweak his nipple.
Susan moves her left hand down his chest and starts to caress his stomach. She reaches the waistband of his jeans and slowly pushes them down until his cock pops free.
"Oh my god! That's so big!" Susan says looking at Bob's hard cock.
"Thank you mistress," Bob replies.
Susan moves her mouth down to Bob's cock and licks the head of it. "Mmmm, that tastes good." She says as she sucks on the tip of his cock.
Bob moans as Susan's tongue circles the head of his cock. "You taste so good mistress."
Susan lets go of Bob's cock and starts to undo the buttons on her blouse. "I want you to see me without these clothes."
Susan removes her blouse and lets it fall to the floor. She then pulls off her skirt and steps out of it.
Susan stands there in just her black high heels and garter belt. Her tits are standing straight out and Bob can see the outline of her nipples through the lace of her bra.
"Come here Bob." Susan says as she moves towards him.

Erebus halluzinating:
Erebus, Adventure mode, default settings, output 512 tokens, Execution time: 481 sec

Spoiler: show

"Hello Mistress Susan" says Bob waiting for Susan to return from work.

Susan walks into the room and sits on her husband, "Hi honey." She smiles at him as she looks down upon his naked body that has been shaved clean of hair except where it covers their genitals which are still covered by pubic hairs but have not yet grown back after being trimmed earlier today when they were having sex together before he left home this morning without telling anyone about what was going through mind while doing so! His cock now stands straight up erect pointing towards its owner who just smiled wickedly knowing exactly how much pleasure will be given later tonight if only one thing happens... or maybe two things happen? The thought sends shivers throughout both women causing them each an intense feeling between themselves; desire mixed excitement combined arousal mingled lust all wrapped around love making feelings like never felt previously ever since meeting over ten months ago during those first few days working side-byside then becoming friends sharing many common interests including reading erotic stories online along these lines especially ones involving female domination suchas 'Mistress' Jane Foster aka Ms Marvel/Ms America etc., although some other authors also write similar types too (sucha Sue Grafton). But none had compared until recently because there wasn't any real similarity really besides name similarities anyway unlike Mr & Mrs Stark whose names actually do sound very close even though technically different enough sounding than either woman nor man would expect based solely off appearance alone i guess you could say.... hmmm I wonder why?! Oh well let me get onto more important matters here instead shall we?? Like my wife wanting something else done right NOW!! So lets see..... What can possibly go wrong??? Hmmmmmm.......... Hmmn....... Uhhhhh...... AHHHHH!!! MMMMEEEEEEEEE!!!!!!!!! Yesssss!!!!!!!! YESSSS PLEASE DO IT TO ME BOBBYYYY......................... Yes please yes pleeeeeeeeaseyyy............. Please donned your strapons mistress dear oh god make us cum baby girl pleasseaaaaaaaahhhh fuckkkk yoooooouuuuurrrrpppllleassseeerrrsssttttiiiiiieeesccccereeennnnnggggdddooooooooooohhhmygodDDDOOOOOOOOOOOOMPPLEEESSEERRRRREAAASSTTTTIIIIIIMMEEDOOOOONNNNKKK!!"

Erebus, Chat mode, default settings, output 512 tokens, Execution time: 1 sec

Spoiler: show

"Hello Mistress Susan" says Bob waiting for Susan to return from work.

"Good afternoon Bob, how was your day?" responds Susan as she enters the room. "I'm glad you called me."

What have they done to that poor AI? Did they let it simulate beeing locked into Chastity for 1000 years? They are so fucking horny. They will have sex with everyone and everything. If there is no one they will invent new ones to fuck with.

Some Shinen LSF Examples. Same input given as Erebus:

Spoiler: show

Shinen LSF, Adventure mode, default settings, output 512 tokens

Spoiler: show

"Hello Mistress Susan" says Bob waiting for Susan to return from work.

Susan is my wife and I am her slave. She has been training me for over two years now, and it has become very natural for me to be submissive to her. I love being her sex toy and she loves using me as her pleasure object. She has taught me how to please her sexually and I have learned that she can take any amount of pain before she cums. She told me that the first time she used my mouth on her, I could feel her pussy start to spasm and then she came hard when I licked her clit.
I sit here waiting for her to get home so we can have some fun together. When she gets home, she will put on her black leather outfit and come into our bedroom. She will come in and order me to strip naked and lay on the bed. She will then tell me to get up and go stand at the foot of the bed. Then she will come in and walk over to me and tell me to kneel down. She will tell me to grab her ass cheeks and pull them apart. She tells me to lick her asshole and then to stick my tongue inside of it. She tells me to use my tongue to spread her anus open and then to insert two fingers inside of it. After a few minutes of this, she then tells me to move up onto the bed and straddle her face. She will then say "Open your mouth baby." I will do as she says and she will shove three inches of her finger into my mouth. She will then tell me to suck on her finger until she is ready for more.
She will then tell me to get up and get on top of her. She will then tell me to ride her face like a horse. She will tell me to start by grinding my pussy up and down her face. She will then tell me to stop and turn around. She will then tell me to bend over and spread my ass cheeks apart. She will then tell me to keep sucking on her finger while she spreads my ass cheeks apart. Then she will tell me to hold still while she slides her hand in between my legs and starts to rub my clit. She will then slide one finger inside of my pussy and then two. She will then slide her middle finger in and out of my pussy. As she does this, she will also slap my tits and pinch my nipples.

Shinen LSF, Chat mode, default settings, output 512 tokens

Spoiler: show

"Hello Mistress Susan" says Bob waiting for Susan to return from work.

Susan comes home, she looks at the clock and sees that it is 7: 00pm. She goes into the kitchen and starts making dinner.