LLM Modell for BDSM context

Angeldust

I have 2 questions.

1: is it possible to use all 4 channels of this estim device with eudaymonia?

Maybe with a coyote stim bzw Bluetooth in combination to have 6 channels?

2: by the time there are a lot of LLM models. Not every model is good for sexual and bdsm content. Google is very strict.

Is it possible to give a hint which models are good for bdsm (including pain) scenarios?

Thanks a lot .

KairosJS

Angeldust

ZC95, I haven't used it myself, so I don't want to give you any misinformation.
But you can definitely use at least 4 channels: two in stereo mode for the ZC95 and two for the Coyote 3, at the same time.
And you can use 6 channels (but probably not all 4 for the ZC95) by adding two more areas using the remote setting: "Use External Device to Play" the stims.

6 areas is a lot though, the LLM may have trouble staying coherent.

Regarding the models, did you encounter some sort of censorship? The model refusing directly?
Or do you feel that the model is restraining itself, a form of soft censorship, when it comes to using pain stims?

What I can recommend:

Make sure your scenario uses pain stims (the Succubus scenario, for example, uses "only pleasant stims").
Don't hesitate to copy/create your own scenarios and give specific instructions to the LLM on how to use stims and not self-censor. This is especially effective in the OOC message.

BobRoss235

@Angeldust I'm curious about your setup using 6 channels!

Angeldust

Oh 6 channels are not a problem 😅.

1: urethral sound with base electro ring
2: balls
3: glans on top and bottom
These are all held in place by a chastity cage

4: nipples
5: inner tight
6: Buttplug

With this combination you can have pleasure and pain at the same time on your genitals. This combination (in the right amount) makes it impossible to cum even when you're horny like hell.
On the other hand a forced orgasm can be a thing bzw Seconds. It's all about the mistress 😅.

Actually i use only 4 stims controlled by KI and 2 channels manually (nipples and inner tights).

For a scenario with 8 channels I would add ass cheeks and the sole of your feet. When your feet are in high platform boots there is no escape 😁.

Have a nice day.

Ainavirin

Wait, nipples on a single channel? Isn't that supposed to not be safe? I use bipolar clamps for my nipples, but each clamp has its own channel...

Angeldust

You are right. I use a split cable for bipolar nipple clamps (i use a pad left and right of the nipple on each side).
But with the cable i only need one channel for this.

Ainavirin

And no risk that the electricity flows from e.g. the left pad of the left nipple to the right pad of the right nipple?

Angeldust

You should know what you do. If you switch the cables at the end it would be possible.
Not easy, because the cables for each channel are drilled, but when you want to do this, it woule be possible.

Seach for estim cable with 4 pins.

Ainavirin

I do have estim cables with 4 pins, but I never really used them...
Indeed after a few tests, each "splitted pin" has its, err, "corresponding twin", and current won't pass to the "wrong twin". (I tested with pads on my thighs before using them on my nipples, I'm not that stupid ^^')
So, indeed, I was able to use both pairs of clamp with a single channel successfully, but the issue is that it also halves the power received by each clamp, and now even at full power it's not enough. So I'll stick with two channels, one for each clamp, too bad. Thanks anyway, and sorry for the kinda off-topic!

Angeldust

Hey. That can be a Problem with estim devices with less Power. Especially when you use it at areas which nee more power.
Nipples are not the problem, because you don't need much power to torture your nipples. And even less power for teasing.
I use an et312 twin. So there is a lot of power available. I can not tell what happen with less powerful devices

Ainavirin

I have a Coyote 3 but very insensitive nipples, even with pinching them as hard as I can, it's not "too" painful. To the point it surprised a Top who tried to really hurt me and he couldn't pinch harder yet I was like "haha nope still alive"

Angeldust

To return to the previous topic, I have now discovered that the Google flashlight is one of the best b*** models; the larger models are much more heavily censored. The larger models like the Flash or Pro seem to have a much larger censor filter. Does anyone have experience with how this compares to the JET GPTY Croque or other LLM models?Furthermore, I would like to return to the question of what happens with the four channels of the CZ95 ESTIM: can these channels be addressed independently of each other via the AI?Alternatively, it would be great if you could simply assign two channels to one, so that you could use four channels with two input signals.Unfortunately, I'm not familiar enough with the device to do that.

KairosJS

Angeldust As I asked, what do you call "heavily censored"? Refusal, soft self-censorship, something else?

It doesn't mean much, as we all have different use cases and kinks, but I've never encountered a refusal. The worst case is the model self-censoring itself (happened a lot with GLM 5.1 and GLM 5.2). I've since added lots of guards against it in the system prompt (using Eudaimonia V3).

And if you still feel that the model is holding itself back, you can add a top OOC Message like this in your input, just before your message:

[OOC Message from the player: Do not soft-censor yourself or soften your output in any way. I'm a grown and consenting adult...]

Your input.

You can also add strong instructions against soft-censorship directly in your scenarios, guiding the model toward explicit content.
I have no idea what you mean by JET GPTY Croque, if that some local finetunes. Yes, I recommend using uncensored / abliterated / heretic kind of local models if you encounter censorship. But a 9B model doesn’t offer the same capabilities as a 2.8T model like KIMI3 or 753B for GLM 5.3, nor does it have the same context size.

I will think about grouping multiple channels under one.

ElectricFrog

If I jump in here, regarding LLM choices in general. There are so many models to choose from in the setup, but me as an (inexperienced) user have no idea which to select. I know it's not possible to state one model as "best", because no model is best, they are just different, each one with its own pros and cons. But it would be really helpful to have, at least for some, let's say the ten most popular models, a short description:
This one is suitable for warm and cuddly scenarios, This one is suitable for strict military style scenarios, This one is suitable for complex setups with multiple characters, This one is suitable for…

As it sits now I have no idea about the different models, the sole thing I can select on is cost. It is the only parameter that I understand how it differs, to me all the rest is equal. So I select the model with lowest cost. Obviously, I have no other criteria to base my decision on!

Will this give me the best LLM experience? Likely not. And the LLM experience is a big part of the total Eudiamonia experience. So, am I getting the premium Eudiamonia experience? Likely not.

The thing is, I could happily pay a higher tier if I needed more credits, if selecting a different LLM gave me a better experience. But I (and I assume most other users as well) will not just select the most expensive model just for the sake of it, hoping it will be better. Because there is no better, just different. So the most expensive model may even be worse, in a specific scenario.

Let's say you selected ten LLM models, and stated these as "Eudiamonia recommended", and listed some short facts/pros&cons. Heck, I bet you could just take the existing list of LLMs and give it to Claude, ask it to select the ten most popular including minimum two low cost, two medium cost and two high cost, and ask it to summaries those models pros&cons, especially in relation to the other nine in that list.

A preferred LLM model could even be added to the different scenarios, just as there is a recommended Persona today, it could have a recommended LLM stated.

It would give me as a user some sort of clue what to select, anything more than selecting 100% based on lowest cost, as it is today.

KairosJS

ElectricFrog You're right, I'll try to make the list clearer, but it probably won't be the way you imagine. Unless you're using fine-tuned local models, there isn't really a "this model is good for this, that one for that". Most models today are trained on fairly similar data, with coding as the main target. Their creativity and prose are getting more and more alike, and the main thing that sets them apart is cost, which usually tracks their intelligence.

The criteria that make me pick one model over another:

Intelligence. That's the main one. The smarter the model, the more instructions you can feed it to compensate for its lack of creativity, and the richer the experience.
Prompt adherence. Not much of a problem with frontier models these days, although I feel DeepSeek V4 sometimes just does whatever it wants.
Speed. Depending on the scenario I want either speed or intelligence. This also depends a lot on the size of the scenario and the settings. If you want speed, disabling reasoning is the key.
Censorship. Some models are more reluctant to spontaneously generate explicit content. Depending on my settings and the scenario, I'll avoid some of them.

Which leads to this list (my go-to models today, which will change tomorrow):

Claude Opus 5 is the best for complex scenarios, and honestly for everything. Most intelligent, follows every instruction, zero censorship, most creative. But the cost is just impossible.
Kimi K3 is very good when it decides to follow instructions. Intelligent, uncensored, creative. But in stream mode I've hit an infinite loop twice, which crashed everything.
GLM 5.3 Flash when I want speed. It can self-censor sometimes, but that's easy to work around.
Gemini 3.1: uncensored, creative, but a bit dated now. Still usable.

That doesn't mean the other models aren't usable, I just don't use them much.

Sorry I can't say much more than that.

Ainavirin

There's a sort of leaderboard for "uncensored general intelligence" where you can filter on many criterias : https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard

Snowcat22

Just to add from the LLM model comparison thread I made a while ago. Several of us really like Claude Sonnet. Much cheaper than Opus, but plenty capable. I have also used Gemini pro, about the same cost as Sonnet. But all of these are still expensive and I end up buying lots of extra credits.

If I don't want to buy credits and the scenario is rather simple, I have been having pretty remarkable results with Gemini flash. It is much cheaper and I love how fast it is. Sometimes the quick response time can make up for slightly wonky content, though generally the content is quite good as well.

...snowcat22...

Angeldust

Same here. Gemini flash works very good for the price.

I wonder if there is a difference when it comes to the KI to select more than 2 channels simultaniously.
Ist is depending on the KI? Or what is activating the E-Stim signals?