NSFW AI chatbots

[Princess Groundhog]

I haven’t used these in a while or checked what new bots might be out there, but Spicychat has been the best in my experience (the closest to Character AI which does the best RP and has the best memory, just R-rated only).

 

[Karminator]

I haven’t used these in a while or checked what new bots might be out there, but Spicychat has been the best in my experience (the closest to Character AI which does the best RP and has the best memory, just R-rated only).

Spicychat just banned step family kink content. Everyone are starting to abandon ship

 

[tontoman]

Spicychat just banned step family kink content. Everyone are starting to abandon ship

Time to switch to the land lady.

--------------------

Yeah in general the memory limits are the biggest problem. Poe with KoboldAI front end was good because it had things like a world setting, things to help your character setup to keep reminding the AI about important things. But KoboldAI stopped supporting POE as POE kept breaking it.
In a way it was a set of invisible direct AI instructions ( ) that I think would auto feed to keep reminding the AI as far as I understood it.

 

[StoneM]

From my experience SillyTavern is the best. But if you don't choose a model it would choose at random a model from AI Horde and then you will get inconsistent results, because every time the answer will be from a different model. Also sometimes it seems the models are token restricted more than their native restriction, because when I run the model on my PC is gives longer conversations, but other times the models from AI Horde work OK. SillyTavern also has very good character description and the characters would not forget their initial description. Of course their short term memory will get deleted after about 2000 or 3000 tokens (depending on the character description).

There is a model with 200 000 token limit, but nobody on AI Horde runs that model, the model is very good but sadly it breaks my PC after sometime, because it needs more than 32GBs of memory, I unload some on my GPU, but even then it overflows, because the longer the conversation the more memory and more time it needs to answer, the model is probably on the level or close to ChatGPT. But sadly I can't use it properly to confirm it in a longer conversation... the model is called Yi-34B-Chat-GGUF if someone is interested to try to run it on their PC, there also different modifications from users on huggingface of this model, but sadly these are even more heavy and I can't run them at all.

From experience I should say "Xwin-MLewd-13B-v0.2" is a good model and always has a lot of horde workers supporting it. I also can run it native and it's one of the best models for NSFW chat. Also some of the 70B models are good but usually there are no people who run these models and the answers are slow.

About sending and recieving pictures from the chatbot, with SillyTavern you can do both... but the problem I have is, I have some bug so half of that functionality doesn't work for some reason and the bot doesn't react properly and is not showing pictures I have to enter specific commands and that breaks immersion.

 

[Daba]

From my experience SillyTavern is the best. But if you don't choose a model it would choose at random a model from AI Horde and then you will get inconsistent results, because every time the answer will be from a different model. Also sometimes it seems the models are token restricted more than their native restriction, because when I run the model on my PC is gives longer conversations, but other times the models from AI Horde work OK. SillyTavern also has very good character description and the characters would not forget their initial description. Of course their short term memory will get deleted after about 2000 or 3000 tokens (depending on the character description).

There is a model with 200 000 token limit, but nobody on AI Horde runs that model, the model is very good but sadly it breaks my PC after sometime, because it needs more than 32GBs of memory, I unload some on my GPU, but even then it overflows, because the longer the conversation the more memory and more time it needs to answer, the model is probably on the level or close to ChatGPT. But sadly I can't use it properly to confirm it in a longer conversation... the model is called Yi-34B-Chat-GGUF if someone is interested to try to run it on their PC, there also different modifications from users on huggingface of this model, but sadly these are even more heavy and I can't run them at all.

From experience I should say "Xwin-MLewd-13B-v0.2" is a good model and always has a lot of horde workers supporting it. I also can run it native and it's one of the best models for NSFW chat. Also some of the 70B models are good but usually there are no people who run these models and the answers are slow.

About sending and recieving pictures from the chatbot, with SillyTavern you can do both... but the problem I have is, I have some bug so half of that functionality doesn't work for some reason and the bot doesn't react properly and is not showing pictures I have to enter specific commands and that breaks immersion.

For this to work, do you need a powerful NVIDIA GPU with at least 10GB VRAM?

 

[Wayne]

What's a good one? one with unlimited chat and no waiting period/or minimal waiting period, and of course NSFW with no filters

thare is none unless you count chai witrh is not time wait based

 

[koiflo]

NSFW AI Chatbots? Nah. I'm still waiting for this: View attachment 2842895

Soon brother

 

[Karminator]

Time to switch to the land lady.

--------------------

Yeah in general the memory limits are the biggest problem. Poe with KoboldAI front end was good because it had things like a world setting, things to help your character setup to keep reminding the AI about important things. But KoboldAI stopped supporting POE as POE kept breaking it.
In a way it was a set of invisible direct AI instructions ( ) that I think would auto feed to keep reminding the AI as far as I understood it.

I heard noval Ai is good , their higher end plan can support 8k tokens vs spicy chat 4k tokens at the same price as.



Do you think it's worth it ?

 

[Karminator]

I've been using various nsfw AI chatbot for 8 days. Here's the verdict: IT'S NOT READY YET.

IMO, it can produce some powerful emotions, but it's highly inconsistent.

Problems
1. inconsistent results
2. forgets important details too quickly, no matter how you program it
3. moves events forward too fast, even if you use code stating other wise
4. deviates from current task as if suffering from attention deficit disorder, despite using code to prevent this
5. if you want to pay $$$, one website's AI might give great responses but lack basic features, while the other website has lots of nice features but a shitty AI response.

AiChat's I've tried
0. kindroid.ai - suggested by user Gardenson - great response, can't edit character message - i'm going to keep my eye on this one and join their discord - i don't like theme of talking to the character via the phone, hopefully we can choose a different theme like a story mode where it feels like you and the character are there in the same area through immersive story telling
1. crushon.ai - amazing response but you can't even delete or edit messages, so what's the point? Their ui looks like shit.
2. aisekai - lots of nice features, can edit, delete, but the ai response is not good
3. yodayo - ai response not good, nice UI, not nearly enough tokens to make simple logic kick in, narration can't be controlled effectively
4. sakura - poor response from what i remember
5. sillytavern +koboldhorde = what a fucking nightmare, hahah, the learning curve is ridiculous with documentation that doesn't cite any examples of usage, so learning it would probably take a whole month.
6. a lot of other ones that were just shitty quality

It has a lot of potential, and I think in about 1 year we're going to see some amazing things from it, but it's just not there yet.

There's also barely any documentation in regards to how to program the bot (you know via {{char}} and {{user}} and all that stuff)

Now, if you guys think I'm wrong, PLEASE prove me wrong, because I want to use NSFW AIChat bot's. When they do actually work, they're a lot of fun, but too inconsistent, still too many problems.

LONG STORY SHORT, if you try to use an AI NSFW Chat bot, you're going to get massively blue balled, because towards the end of the conversation, the AI will forget all the important details when it matters most, the CLIMAX, so you get fucked over every time.

This is very true , I try to get summary of the chat from the we bot and past it again in new chat after few 3k tokens. That way I can progress the plot as long as I want .



Have you tried novel ai or spicychat? What's your opinion on these? Are these under your "shit" category?

 

[hakarlman]

This is very true , I try to get summary of the chat from the we bot and past it again in new chat after few 3k tokens. That way I can progress the plot as long as I want .



Have you tried novel ai or spicychat? What's your opinion on these? Are these under your "shit" category?

What nsfw ai chatbot do you use to include that summary?

From what I hear, spicychat has a harsh filter, so I won't even try it, don't have the patience to be constantly filtered. (I still haven't drank coffee yet, so I may be wrong, but I think it has a filter).

I tried NovelAI but it doesn't feel like I'm talking to a person. However, from what I've researched, NovelAI are making a separate service where you can create a character and talk to them, I forgot what they called the website, it starts with an A, maybe someone else here knows? I've been wanting to try it. Aethei, Aesta, something like that, I forget.

 

[StoneM]

For this to work, do you need a powerful NVIDIA GPU with at least 10GB VRAM?

You can configure SillyTavern to use koboldhorde, which is free.

I use koboldhorde (AI horde), for text generation and local Stable Diffusion for image generation, but still have some problems on making the whole thing working seamlessly I mean (text + image), for text everything is working OK and is probably better than most services. Also I use koboldhorde to run some models which I can't run on my GPU (or are very slow on my GPU).

SillyTavern has lot of configurations which might confuse some people, who are not experienced with local installations. The good thing about it, you can use multiple models inside, with character cards from almost any other chat bot. Also SillyTavern has a sumarry function which works on a CPU (and works pretty well), for then making images on Stable Diffusion. Sadly I have a problem running it seamlessly but it works very well (when it works). Also your bots can show moods, with pictures and things like that.

Probably the simplest thing is to use SillyTavern and run koboldhorde models from the net. Because it's free and it's probably the most easier configuration.
Runinng local models with SillyTavern, needs oobabooga installed and configured... and then it uses API to run the models installed on the oobabooga. SillyTavern is better UI for the models, than oobabooga itself, thats how I've found it.

By the way for everyone who wonders, you don't need a fast PC to install and run SillyTavern itself. It's just a UI with nice features, the models themselves can be run from the Internet and it's not hard to configure that, there are tutorial videos. At least with SillyTavern you can check a lot of models from koboldhorde for free.
Most fancy "NSFW" models which need subscription, only give you a few words for free, before you have to pay money... and usually they are not impressive and most of them are censored, or get censored in some way about a month or two after you subscribe.

One more thing I almost forgot (I'm not sure if it's mentioned in previous comments). You can download character cards and lore books at chub.ai, these are compatible with SillyTavern and other UIs and services.

 

[hakarlman]

You can configure SillyTavern to use koboldhorde, which is free.

I use koboldhorde (AI horde), for text generation and local Stable Diffusion for image generation, but still have some problems on making the whole thing working seamlessly I mean (text + image), for text everything is working OK and is probably better than most services. Also I use koboldhorde to run some models which I can't run on my GPU (or are very slow on my GPU).

SillyTavern has lot of configurations which might confuse some people, who are not experienced with local installations. The good thing about it, you can use multiple models inside, with character cards from almost any other chat bot. Also SillyTavern has a sumarry function which works on a CPU (and works pretty well), for then making images on Stable Diffusion. Sadly I have a problem running it seamlessly but it works very well (when it works). Also your bots can show moods, with pictures and things like that.

Probably the simplest thing is to use SillyTavern and run koboldhorde models from the net. Because it's free and it's probably the most easier configuration.
Runinng local models with SillyTavern, needs oobabooga installed and configured... and then it uses API to run the models installed on the oobabooga. SillyTavern is better UI for the models, than oobabooga itself, thats how I've found it.

By the way for everyone who wonders, you don't need a fast PC to install and run SillyTavern itself. It's just a UI with nice features, the models themselves can be run from the Internet and it's not hard to configure that, there are tutorial videos. At least with SillyTavern you can check a lot of models from koboldhorde for free.
Most fancy "NSFW" models which need subscription, only give you a few words for free, before you have to pay money... and usually they are not impressive and most of them are censored, or get censored in some way about a month or two after you subscribe.

One more thing I almost forgot (I'm not sure if it's mentioned in previous comments). You can download character cards and lore books at chub.ai, these are compatible with SillyTavern and other UIs and services.

You don't happen to know of a guide that can crash course someone to use SillyTavern & KoboldHordeAI? I have it installed successfully, but it has a high learning curve which is time consuming. I read over the documentation yet it doesn't cite examples of usage, so learning it requires too much trial and error, and an excessive amount of research that doesn't yield good results.

Also, do you find SillyTavern & KoboldHordeAI to give consistent responses or do you feel it gives wildly random and jarring responses usually? A lot of the text response also has weird symbols and formatting that doesn't look visually appealing, but there might be a way to clean that up?

 

[StoneM]

You don't happen to know of a guide that can crash course someone to use SillyTavern & KoboldHordeAI? I have it installed successfully, but it has a high learning curve which is time consuming. I read over the documentation yet it doesn't cite examples of usage, so learning it requires too much trial and error, and an excessive amount of research that doesn't yield good results.

Also, do you find SillyTavern & KoboldHordeAI to give consistent responses or do you feel it gives wildly random and jarring responses usually? A lot of the text response also has weird symbols and formatting that doesn't look visually appealing, but there might be a way to clean that up?

When you connect to KoboldHordeAI, you have to choose a single model in the Models tab (it's just below the API Key tab) otherwise it would use different model at any response and it would be inconsistent conversation. I think Xwin-MLewd-13B-v0.2 is the most versatile and is also relatively fast.

Also in the tab which is like a big A , there is Context Template, I put ChatML in there. You can also enable or disable Instruct mode and put ChatML there as well.

At the far left tab in Kobold Presets I use RecoveredRuins as my preset, but you can experiment with different ones.

As for a comprehensive guide for SillyTavern, there is not such thing, because SillyTavern has too many options, but there are a lot of videos with guides on YouTube. When I've started I used this one:

You must be registered to see the links

Then I've mostly researched by myself and had read the documentation.

About the weird symbols. Probably a wrong formatting in the big A tab (it's called AI response formatting), there is something wrong there and that's why the punctuation is wrong.

Maybe if you show some example chat and tell me which model is doing the said strange formatting I might be able to help.

 

[tontoman]

What nsfw ai chatbot do you use to include that summary?

From what I hear, spicychat has a harsh filter, so I won't even try it, don't have the patience to be constantly filtered. (I still haven't drank coffee yet, so I may be wrong, but I think it has a filter).

I tried NovelAI but it doesn't feel like I'm talking to a person. However, from what I've researched, NovelAI are making a separate service where you can create a character and talk to them, I forgot what they called the website, it starts with an A, maybe someone else here knows? I've been wanting to try it. Aethei, Aesta, something like that, I forget.

Yeah think you might be mistaken with Spicychat and thinking about another. I don't notice any filters, although hard to say if something is just not there. But certainly haven't hit the CAI *NOPE* popups or the cutoff aborted responces. And I've seen full on sex, kinks etc. My only complaint with Spicy is it seems to have a worse memory limit, even if you try and keep the character description smaller.

I heard noval Ai is good , their higher end plan can support 8k tokens vs spicy chat 4k tokens at the same price as.

Do you think it's worth it ?

Never tried NovelAI so can't say. And never sure if those limits are adjustable for character definition and/or memory size.
---------------------------

As memory is always going to be a limit on longer chats (if you're going for a long story vs a scene), I'd put more hope into what KOBOLD and POE had with things like world and lore injections. That ways it would 'remind' the AI of certain points automatically while you just do the chat/actions. Because no matter how much mem you have, it will eventually scroll off the chat.

 

[tontoman]

BTW if you don't mind waiting and have a decent CPU, you can run AIs on CPUs. Think when I last tried it it was around 40seconds to a minute for a 32B model on a 5600x, 32GB ram. Which is kind of sad when you think all the work done just for a few sentences.
Uses Pygmalion to do so iirc. Been a while so can't exactly remember. No links on how to (just search Pygmalion) as probably gone or totally different as AI stuff seems to change weekly anyway heh.

 

[StoneM]

By the way Spicychat putting additional censorship is bad, very bad. Any time any form of censorship is put on bots, they become more lobotomized. So I really don't understand all that strive for censorship.
You can try that on your own bot, put any kind of censorship (that's actually done by putting a prompt in the character card)... and you'll see how the bot will become "braindead" much faster. That's why all the bots brake down like that, they try so hard to "adhere" to the censorship, they're becoming like zombies. That means you will be cut off with the usual "nonsense" the moment the bot suspects you're doing something which is not allowed (even if you're not actually doing such a thing).
I was considering trying Spicychat, but the first "thing" I see on their site is about adding more censorship, which is a big no no for me, even though I rarely (if even) engage in said conversations with bots. (Siblings fucking each other (related or unrelated) is not my fetish, yet I strongly oppose any censorship, especially in private conversations, because it's a slippery slope. We've been there already in the middle ages, and it's not nice to repeat this mistake again. Freedom of though is important, also adult people should make a difference between fantasy scenarios and reality.)

 

[tontoman]

By the way Spicychat putting additional censorship is bad, very bad. Any time any form of censorship is put on bots, they become more lobotomized. So I really don't understand all that strive for censorship.
You can try that on your own bot, put any kind of censorship (that's actually done by putting a prompt in the character card)... and you'll see how the bot will become "braindead" much faster. That's why all the bots brake down like that, they try so hard to "adhere" to the censorship, they're becoming like zombies. That means you will be cut off with the usual "nonsense" the moment the bot suspects you're doing something which is not allowed (even if you're not actually doing such a thing).
I was considering trying Spicychat, but the first "thing" I see on their site is about adding more censorship, which is a big no no for me, even though I rarely (if even) engage in said conversations with bots. (Siblings fucking each other (related or unrelated) is not my fetish, yet I strongly oppose any censorship, especially in private conversations, because it's a slippery slope. We've been there already in the middle ages, and it's not nice to repeat this mistake again. Freedom of though is important, also adult people should make a difference between fantasy scenarios and reality.)

You do realize Spiceychat is free? It's not like you're going to be locked into paying for something that's going to change. So there's literally zero cost to try it to see if you like it. If anything, using a throwaway e-mail to join and cancelling later it if they change, does even more to tell them they shouldn't censor. As it will register they lost a potential customer in their numbers. So not really understanding your logic about not even trying it.

It's not like Ubisoft, EA etc. "live service" games. Or Sony and movies thanks to Warner Bros, where you actually lose money spent.

 

[SharkVampire]

You must be registered to see the links

Pretty good attempt by one of the users on the kobold AI discord server to let people vote for the best models and services with chatbots. No registration needed, you can leave your own feedback.

 

[Daba]

By the way Spicychat putting additional censorship is bad, very bad. Any time any form of censorship is put on bots, they become more lobotomized. So I really don't understand all that strive for censorship.
You can try that on your own bot, put any kind of censorship (that's actually done by putting a prompt in the character card)... and you'll see how the bot will become "braindead" much faster. That's why all the bots brake down like that, they try so hard to "adhere" to the censorship, they're becoming like zombies. That means you will be cut off with the usual "nonsense" the moment the bot suspects you're doing something which is not allowed (even if you're not actually doing such a thing).
I was considering trying Spicychat, but the first "thing" I see on their site is about adding more censorship, which is a big no no for me, even though I rarely (if even) engage in said conversations with bots. (Siblings fucking each other (related or unrelated) is not my fetish, yet I strongly oppose any censorship, especially in private conversations, because it's a slippery slope. We've been there already in the middle ages, and it's not nice to repeat this mistake again. Freedom of though is important, also adult people should make a difference between fantasy scenarios and reality.)

Spicychat is actually still very much NSFW friendly, but I've read that they banned incest. Since that is not my kink, I didn't really notice any change, but I despise any form of censorship on principal level. For free service, Spicychat might still be the best option.

 

[Urosan]

I've given a couple of AI chat sites a go, both with released stuff and making up my own scenarios.

Spicychat:

• There's a queue before you can begin, but then no maximum session length. Fine.
• Can generate avatars for characters
• Character creation guide is okay, not great
• Login is all email codes, slightly annoying
• You can't see other character's tags, so can get sprung with stuff you aren't looking for (futa, 'you submitted and now I'm an uwu cuddlemonster', injury)
• As conversations continue it did seem to forget premises, and started just repeating phrases/specific wordings, and different characters shared the same phrases
• You can edit past messages, but cannot revert.
• You can AI-fill in player responses, if you want to drive the situation as the AI understands it

Janitor AI (default settings):

• No queue
• You can check the tags on other's creations
• The creation guide seems better
• Every conversation has a time limit from Delayed Onset Verbosity. As you get deeper into a chat, descriptions increase in length and get more esoteric and metaphorical, until it isn't actually parsable
• You can edit your and AI messages, and can revert to an earlier point

From what I'm seeing, the best free experience (higher token limit > character can hold more tags and remember more things) is SillyTarvern and KoboldHordeAI?