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

[kinkylucy]

There are so many hacky solutions to getting your AI to control your toys, but what if it could control your toys directly? While generating images and showing you porn?

This is Ethumia. A web based app for AI Sex & Kink.
Play now: https://ethumia.com




Ethumia is an all-in-one platform for AI-driven webteases. Instead of duct-taping together five different tools char.ai with xtoys etc, you get one interface where your AI characters can actually do things — control your toys, generate images of themselves, remember your conversations, and guide you through complex branching scenarios.

Into CBT? Pain is also an option.
Fully customisable characters, fully customisable scenarios, but easy to get started!

What it does:
- Sex Toy Integration — Buttplug.io/Intiface, Lovense, Coyote V2/V3, OpenShock, Restim, Audio Estim, and direct Bluetooth support for 100+ toys (chrome/edge).
- AI Characters with Memory — Your characters remember past sessions and grow with you
- Image Generation — Local via ComfyUI/Automatic1111, or cloud-based (limited) for those without beefy GPUs or who want to dip their toes in.
- Scenarios — Pre-built experiences, or build your own with the simple scenario builder
- Kink Support — Hypnosis triggers, Chaster integration for chastity, customizable stimulation and soon, Exhibitionism... Your AI will be able to see you!
- Voice — Experimental TTS voices for characters, voice control for ahem, hands-free use
- AI Tools — The AI can search for and display porn during chat

The setup:
- Free accounts get 200 LLM generations/month on Ethumia Cloud.
- Run your own models on your PC for unlimited. <-- Takes 20 minutes and a decent PC.
- Or pay for unlimited. <- Also an option, if you want it.

It's not perfect — there are bugs, and some features are experimental. But it's actively developed and the 1.2 release just dropped with major UX improvements. It's improved a lot since first release!

Play now: https://ethumia.com

--

Spoiler: show

New account on account of my gender transition and old account having deadname lol.

[verynicekojak]

It would be great to have modern alternative for EOS. The more the merier. But at the moment I didn't manage to start a tease.It is already working?

[thebears73]

after 35mins of trying to make this work. I gave up

[Bumbum]

I have also no success. I fail already at the first step by configuring the AI api. I already have several openai api keys for other applications, so I think I know what I am doing. But with the Base URL https://api.openai.com/v1 I always get an Test Error: NetworkError when attempting to fetch resource.

[buggy82]

Summary/Praise:
If this had an easy onboarding experience, this might become an awesome AI estim platform with unlimited shareable user scenarios and characters. Could even become commercially viable for the creator.
It lacks at least one free preconfigured easy-to access scenario for trial.


User Feedback/Criticism:
As it is right now, you need to be an AI expert and spend money to even get it running for a test:

• For each scenario, you need to create different "characters" first, which are somehow limited to a maximum of 500MB "download" (?!).
  You need some kind of LoRA files for every character, seems to be some kind of character definition files.
  You need an paid API key to an AI API.

The UI is still pretty clunky (e.g. clicking on a tile does nothing. Does not open the information/detail page. You HAVE to click the "Info button" inside the tile)
some features don´t work (e.g. LoRA file download leads to JSON error in line 1). All of this is to be expected in such a new project and can easily be fixed.

There are so many different settings pages with so many settings, options and technical AI terms, even the most motivated (horny) e-stimmer without a Bachelor of science in AI will probably give up before getting the first session to run.

I see a lot of potential in this project and wish the creator all the best.

[thebears73]

I have also no success. I fail already at the first step by configuring the AI api. I already have several openai api keys for other applications, so I think I know what I am doing. But with the Base URL https://api.openai.com/v1 I always get an Test Error: NetworkError when attempting to fetch resource.

same issue for me, wasted way to long on it.

[kinkylucy]

The issue with OpenAI Compatible LLMs has been fixed.
I've been working overtime to fix bugs on this so all your feedback is appreciated.

Changes made in the last 24 hours
- Added advice on Ollama CORS
- Fixed character preview generation (no sampler issue) -- thanks @Spejs0nx for the report.
- Added discord bot which posts #new-content and #new-characters
- Add discord connection in User Profile for @member Role
- Added favicon
- Fix OpenAI Compatible APIs
- Fix visual issue with dropdown menus
- Fix forums not showing profile pictures
- Various other bug fixes and visual fixes.


It is in alpha so I would encourage you to come back in a week if you had issues the first round as I hammer out all the edge cases and make it smoother.

The 500mb daily download limit is for files/LoRAs. You can also download LoRAs at civitai.com.

[MarsRaven]

I was able to successfully install and connect Ollama and a model (you had good instructions).

But I cannot for the life of me figure out how to use one of the downloaded characters. Nor can I find instructions.

[kinkylucy]

I'll be releasing a setup wizard this weekend which will take you through the entire process.

For the characters, you need to get a model from civitai.com like Pony Diffusion or Illustrious and put it in the comfyui / models/checkpoints folder. Then download a character lora from civitai or our site and put it in models/loras and then on the characters page you can create a new one, prompt should be something like "1girl, solo, blonde, glasses, and any prompt the lora tells you to put in there." negative prompt should be something like "2girls, multiple girls" and any negative prompt the model says on its page to put in there. Then scroll down to Character lora's and click that, click something like "discover from comfyui" and your lora you downloaded and put in /loras will be there. Add it, and generate your test image, adjust the prompt if you want to change things, and then you're done -- save it.

Btw best model for ollama is: ollama pull deepseek-r1

[kinkylucy]

I've just released 0.2, this release has major improvements in stability and setup. As well as experimental support for almost all common sex toys!

Onboarding - Onboarding Wizard to get you started faster!

Scenarios - New JOI Scenario with No AI (as an example of a no AI scenario): https://ethumia.com/community/scenario/ ... 4485112bc7

Sex Toy Support
* Restim Support (FOC-Stim, Audio Estim alternative)
* Buttplug-js support (Intiface, sex toys)
* Lovense Experimental Support
* Experimental Coyote Support V2 & V3
* Experimental Openshock Support

AI Features
* Allow for disabling of LLM
* Allow for disabling of Image Gen (reduces load on LLM too)
* Your characters now have memories
* Added Numbered Buttons as suggested by @_chrisbeck in #-suggestions
* Star images, use them when imagegen isnt avaliable/disabled.
* Password reset
* More reliable loading of scenarios
* Loading bar for loading scenarios
* No AI Scenarios
* Discord welcome
* Manage memories
* faster loading
* Better SEO
* Many other smaller improvements
* Many scenario bugfixes
* Many, many, many, many, bugs fixed.

[Tiri]

is it possible to add chutes.ai support like openrouter? I can't get Ollama to work properly.

[kinkylucy]

If it exposes a openai compatible api you should be able to use it now, with the "Custom API" setting. Otherwise I can suss out chutes.ai over the weekend.

[Bumbum]

I am a step further, but my PC seems not to be powerful enough to run ComfyUI. The generation always times out. This is my first AI own-host experience and it seems like I run out of memory ( 8GB GPU; 16 GB RAM seems not enough).

Is there an alternative?

[kinkylucy]

Is there an alternative?

8GB is going to be tight if you're on an older card or running ollama. Try making your generation sizes smaller 512x1024 and reducing the steps to like 15.

You can turn off Image Generation. I will have cloud image generation at some point but it's expensive to offer, so will be a tier above supporter+.

[Tiri]

If it exposes a openai compatible api you should be able to use it now, with the "Custom API" setting. Otherwise I can suss out chutes.ai over the weekend.

I made it work in "Custom API". It wasn't an option in the introductory setup, so that had me confused for a second.