[Release] ChatGPT + Sex Toys = Ethumia 1.2: Webtease with AI

[kinkylucy]

To be on the safe side, I would like to mention that you should not misunderstand my answers. I think it's brilliant what you've created and I just want to give you feedback on my experiences. If something comes across as negative, it may be due to the translation. My mother tongue is not English.

I completely understand, I appreciate the feedback -- it's so useful for making the system better.

I've just released an update that should improve the quality and speed of the free LLM. The alternative is to install ollama locally (if you have a GPU with extra headroom) or use openrouter and deepseek-v3.

[kinkylucy]

The image generator, i managed to set it up to at least be working, but i am having distorced and bad images all the time, so i have no idea how to set it up to obtain decently looking images.

The update to the LLM should fix the issues you've been seeing.

Also if you go to Comfy UI settings in Settings -> AI Settings -> comfy UI comfg and change the scheduler and sampler. https://ethumia.com/settings/ai/comfyui

Try:
Sampler:
dpmpp_2m
dpmpp_2m_sde

Scheduler:
simple (if having issues with...)
karras

Change one at a time and press "Test Generate Current Character" and change it until you're satisfied.

Increase steps if characters are weird, will take longer to generate but will be better.

Unfortunately it's really hard to give good defaults since it relies on the GPU you have and the model you've selected.

[Bumbum]

The alternative is to install ollama locally (if you have a GPU with extra headroom) or use openrouter and deepseek-v3.

I am the guy with the very weak GPU (2060 Super) with only 8 GB of RAM. This would not be good for a local LLM. The image generation works now, but with your recommended settings this not good either. The generated pictures of the characters change face every generated picture for example.

I think the long wait times is what turns me off the most. The response times of the Free build-in LLM are at least 30 seconds, but sometimes 1 to 2 minutes. And then there's the additional time it takes to generate the images. In our short-lifespan-trained life this is killing the vibe.
But I see great potential here. If the answers come very quickly (let's say 5 to 10 seconds) and then take that long again for a cleanly generated image, you've created something great.
Could it be that you are not using the "streaming" function of the LLMs? As it looks to me, you wait until the complete response is generated by the LLM and only then forward it to the tease as a whole, not token by token as you are used to?

What I also really missed during my session this morning was some kind of autoplay. I started with "Play" and no scenario. I explained my scenario in my first request to the character (LLM). That worked quite well, but I always had to type something to get it to continue. Sometimes it would have been nice to tick "Auto Play" with an adjustable time. After the time has elapsed, you could internally instruct the LLM to continue in the story, generate a new image and then adjust the stimulation accordingly.

[kinkylucy]

Could it be that you are not using the "streaming" function of the LLMs? As it looks to me, you wait until the complete response is generated by the LLM and only then forward it to the tease as a whole, not token by token as you are used to?

Unfortunately yes, because the LLM does multiple things like pick clothing, take actions, etc. Using the streaming function isn't practical.

Right now there are more retries happening than expected, especially in longer chats. Meaning it can take 1 minute to get a response that is valid and show it to the user. I'm working on making it more reliable, and therefore faster, with the hope to get it under 15 seconds everytime.

I'll look into doing an autoplay feature soon, at the moment I am focused on making existing features faster, more reliable and better.

[G3General]

I set this up properly yesterday and had a serious go. Here is my two cents.

What i got working: Integration with ReStim. Integration with comfyUI. Ordinary chat.
What dosn't work: Scenarios.

Took me about 2 hours to set up all the parts from scratch, including installing comfyUI, Git, downloading models, creating characters and so on. Started a couple of scenarios, but for me they are all broken. "Normal" AI chat with just one AI works though. Response time is great at start of a new chat, but diminishes noticably after 15min. Here, response times becomes unacceptable, as others have mentioned.

My PC with comfyUI generates pictures on avrage in 28 seconds, which is to much, but this is basicly not a tech problem but a wallet probelm (my wallet :p) .

All in all I think this is working quite nicely. It is unfortunatly in the current state not going to be used a whole lot, mainly because of the delay in AI chat. Some delay is fine. 15 sec is fine for chat, but for this to work with sex toy integration, it will have to work immdiatly after you for example say "on the edge", "about to cum" and so forth.

[kinkylucy]

All in all I think this is working quite nicely. It is unfortunatly in the current state not going to be used a whole lot, mainly because of the delay in AI chat. Some delay is fine. 15 sec is fine for chat, but for this to work with sex toy integration, it will have to work immdiatly after you for example say "on the edge", "about to cum" and so forth.

I'll be looking into issues with scenarios. Sorry you had issues there.

I will add an "I'm edging" button that will stop all stimulation and send that as a message on click, as a compromise, on the next update due next week.

I'm definitely building this for a future where users can run models that are much faster on their home PC's.

[fossie67]

got this running with restim , spent 2 hrs playing Ai Dominatrix was fun had me leaking precum . However needs more to get you to cum i think , tried asking begging but no release .

[Iambadwithnames]

Quite the interesting site. I appreciate the ambition with all the different systems you offer integration for.
I did notice that the dropdown to select a local ollama model doesn't really work. I click on a model name and it doesn't save the input. And when you try to enter the model manually, the input box deletes any character as soon as you enter it.

[kinkylucy]

Quite the interesting site. I appreciate the ambition with all the different systems you offer integration for.
I did notice that the dropdown to select a local ollama model doesn't really work. I click on a model name and it doesn't save the input. And when you try to enter the model manually, the input box deletes any character as soon as you enter it.

Thanks for feedback, I broke this when upgrading a different system -- it'll be fixed this week in an update. Apologies!

[splentis]

This has huge potential, and I’m impressed with what you’ve achieved so far.

I’m fortunate to have a high-end rig, so SDXL and Deepseek run quickly—under 10 seconds. Waiting more than 15 seconds would likely break immersion and test patience, though that’s very dependent on player circumstances and largely outside your control.

At the moment, the scenarios don’t quite work. They often double up on LLM prompts, get confused about who is doing/wearing/asking what, or throw errors like “player trying to move object” or “changescene.” The generic chatbot is fun, but it also shows its limits. Some hardcoded player state buttons could help address these issues (stop/slower/faster/edge), as well as letting the bot auto-progress along a theme, with the occasional prompt to steer it in a direction.

A wishlist or roadmap on the forum might also be useful.

Keep up the great work!

[ILuvPlanes208]

I send you a PM with an issue I am having getting this to work. Thanks!

[fossie67]

not working right in ai dom since you made changes

[fossie67]

Still not working as it was before you changed things > Ai dom when playing repeats commands and restim signals are hit and miss . Hope the OP gets back about this but seems not to reply here ?

[kinkylucy]

All the issues above have hopefully been resolved with the latest release!

HIGHLIGHTS
Ethumia Cloud Image Generation!
Simple image generation for free, supporter and supporter+ users.

Better Image Generation with better posing!
Huge improvements, basically a night and day change. Let me know your feedback!

New Simple Scenario Builder!
No more putting scenario text into your characters or starting off a chat with a prompt! More scenarios coming soon, and build your own now!

New experimental Direct Bluetooth Toy Mode!
Connect your Lovense, etc, directly via Bluetooth (Chrome/Edge only)

New EXPERIMENTAL TTS Voice AI Models!
Experimental AI Voices for your characters!

New AI Tools!
The AI can begin to control things... first up, now it can look for porn and display it to you!

Improvements
- UX/UI Greatly Improved
- Auto Respond mode greatly improved.
- Automatic1111 Initial Support
- Chat UI Improved
- AI Memories better than ever.
- You can now duplicate your characters to make variations easier.
- Sharing your characters is easier than ever!

Fixes
- Further fixes for Audio, Restim and Coyote.
- Setup Flow Improved Greatly

STOP/Safeword now always avlaiable, even when stim stopped.
- Removals
- Removed Lovense in favour of bluetooth mode.
+ So much more!

What's next? Better sensations through your toys with more customisability and better base packs for restim, audio, coyote and bluetooth toys. I hope you all give it a chance!

[markus]

No idea whats wrong.
Also the error msg is there just for a very short moment after pressing test, had to screen record to get that screen.

Not in the mood to do restarts or checking console outputs, too often such toy-control apps simply doesn't work, sorry, it's frustrating.

ethuDingens.jpg (198.11 KiB) Viewed 3397 times