EstimAI Roadmap & Feature's requests

KairosJS

Current Roadmap, WIP & Ideas

Below are some important features that I've started working on, or that I'm envisioning for EstimAI. I'm not sure they'll all see the light of day, but here's where most of my current and upcoming efforts are going:

Large Features:

🟢 More modules:
- Giving the AI more tools to play with you, especially through function calling. For example, a few simple modules come to mind: Random (dice, wheel, cards, etc.) / asking the user a direct question or a dilemma that requires an immediate, timed answer.
🔵 Sound FX?
- Like the Screen FX Module: create a sound FX agent that would sync audio cues to segments. This might be really useful for some scenarios. I'm wondering if it should also add some expressive sound cues, such as moaning, laughing, etc.
🟡 Voice detection:
- A simple module that the AI can use to monitor volume/voice activity, calling back to the model if something is detected.
🟡 FX with intelligent analysis of the currently displayed picture:
- This should allow for better camera movements, for example: zooming in on a specific body area.
🔴 Animated Avatars:
- Self-explanatory. I started developing this a few months ago, but it takes so much work and goes against the "Intelligent Image Generation" feature that I love so much, so I decided to stop. That said, it's still on my mind.
🟢 New game mode:
- This is my main priority, but it requires a lot of work. A completely new game mode: a very immersive and personalized simulation of a BDSM relationship between the user (as the sub) and the AI (as the domme) over a long period of time (2 weeks?).
🟠Using Stream for faster / immediate generation:
- This could be a game changer for many scenarios (especially regarding edging / ruining, etc.), but there are a lot of issues tied to it, since EstimAI was never built around it.
🟠 Position Tracker:
- A position detector that would allow for a "Virtual Bondage" feature. The idea is simple, but it may seem incredibly intrusive: it would use your webcam to take pictures of you every few seconds. The photos would, obviously, never be sent or stored online, but would be processed by a small AI in the browser that interprets what position the user is in. It can then call back to the AI to tell user's position, or if he moved for example.
🟠 A Biometric Module (Edge detector): as dystopian as its name sounds, this would be live biometric monitoring that measures the user's level of arousal for edge detection. The data would be passed to the AI. This would be really immersive for certain scenarios, but it has some important drawbacks:
- It's intrusive and dystopian as hell.
- It requires the user to buy hardware (probably a Polar H10).
- It won't be 100% accurate and will require calibration.

🟢 In progress
🔵 Worth the time?
🟡 Planned
🟠 Planned, but...
🔴 Aborted for now

Smaller Features:

Audio output selector for Stereo Stim
Better phase editor and handling for scenarios
Release of Stereo V2
---

Your Requests

Please write a clear and structured post for your idea / feature:

The purpose of the feature
Your suggested solution (workflow, tech)

I read every single request and am eager to integrate them. But time is the main constraint, so unfortunately I can't promise I'll build them all.

So please feel free to comment on the roadmap as well, so I can get a better understanding of what I should and shouldn't put my time and effort into.

Tycho

Feature Requests:
-Persona thumbnails for personas similar to the feature for Scenarios.
-Dedicated section for personas to facilitate users generation of images consistent with each character. Like AI Character Generator presets that can be toggled when selecting the scenario.

Welwa

I have elements that i include in some scenarios that i would like to use in other scenarios but not all the time. so a feature where you create blocks of instructions that can be added to the scenario before play begins would be great. it would be like the generation enhancer in the new modules section, with a list of your blocks than you can add to the scenario on the fly.

Mesh1401

Feature Request:

to have an option in Normal Mode to freely specify dress level for the 3 stages of the Normal Mode tease and enforce that to AI image generation (or galleries). Currently you have added an option to 'force specific nudity level at start' which is great and many thanks, but the tease always ends 'fully naked', which is probably good for most normal people but I am -unfortunately(?)- not normal: would very much prefer still half-dressed ;-)
I understand it could be too difficult to implement due to interference with existing dialogs and mechanisms, so just a friendly humble request :-)

Ainavirin

Feature request:
In the Characters tab, allow to pre-generate an entire gallery for a character (limited to 10 images for free users, more for the others), along with their description.

Why:

It would allow to disable image generation when running E-stim AI so that we can instead use a predefined gallery (I can't run both image generation and a LLM locally).
It would allow to use images that took a long time to generate (so, high quality images) and to carefully select them, instead of having to ensure the generation is fast enough to be used in real time with the added risk of having bad pictures (extra limbs or other visual glitch)

fourier

Feature request:
Support for alternate LLM providers such as OpenRouter. This should be an easy addition, as it is largely the same as for LM Studio or other local options, except the API key needs to be included in the request. This would allow users to make use of the free models that are often available that are larger than they are capable of running on their own device, or provide an alternate paid option.

jp774

Feature request: Ability to import audio signals for stim.

While the Estim signal editor is nice, other tools have more feature for signal creation, and there are a hundreds of stereo stim signals that are already out there, and it would be nice to have the ability to use them on the platform.
I dont know all of the constraints you have, but if hosting a lost of sound files is too heavy, i imagine that it could be done in a similar way as when we use our own images and videos in Eudaimonia mode. As in, you dont actually have to upload everythings, but you give access to a certain folder which contains the relevant files. You could ask for a different folder for each type of signal (tease, edge, pain, orgasm) in order to simplify identifying the purpose of each signals. Even if signals have to be restricted to a certain length, that would be quite a nice feature.

Edit: By the way, im very curious about what is Stereo V2 going to be like.

Banham

Feature Request
You were thinking about A Biometric Module .
How about implanting the Pawprints from DG labs as you already have the bluetooth interface to run the coyote V3 ?
They have schock/tilt sensors and also a integrated A/D for external sensors , would be ideal for edge detection in the tease denial sessions ..... Could also be used as answer buttons when you are not able to type .

Snowcat22

Banham
Love this! I bought the Pawprints when I bought my Coyote and haven't even opened the package yet because I'm not sure what to do with them. Would love to try them out!

Tycho

Feature Request: Ability to set "Message duration coefficient" below 0.7, including to 0 without being in debug mode.

Justification: When testing things I don't want to have to wait for each line to be displayed or use debug mode before I can interact again or edit the whole message. Sometimes, I prefer if the whole response was immediately spit out, ready for editing or my next response.

Workflow: This would probably result in skipping over stims if there are several in one response. This behavior is expected. Perhaps this mode could just default to the last stim in a message or play through the stims at normal speed independently after the whole response is displayed.

BobRoss235

Feature Request:
Adding an option that increases the estim limit by X every Y minutes.

Justification: I typically set the "absolute limit" of a channel a bit higher and then start with the "scenario limit" a bit lower. Over time I increase it manually via the scenario settings. It would be great if this happens automatically based on the scenario timer. It would be required for both normal and pain values, eg. increasing by "1" every "5 minutes".

MalteP

Its great that we now have so many toy options with all the new integrations but I always felt that a natural extension to this would also be the option to just show a beatmeter. I mean wouldnt it be easy to abstract one for each pattern ? It could just be another toy added in the controls. Same for like ball slapping. like a secondary beat meter with a way of saying how hard. I know this is not the main purpose of the website but even for just testing out scenarios this is a great use case.

raider

Import/Export function for the estim editor (Coyote). It seems like the .pulse file format or various json formats are firming up for the Coyote right now. If this can be used there is a lot that can be imported in the library.
Ability to view and generate revisions of community stims in the estim editor

Alpako

Feature Request: Krea 2 generation option. Its awesome model!

KairosJS

Alpako I've just added a "krea2" option for Clip model type in eudaimonia. You should be able to use krea2 in eudaimonia by using the "Diffusion models" template.

From my tests quality is decent with this. Make sure to use one of the following to uncensor it:

A filterbypass
A NSFW lora
A uncensored checkpoint

I've been using Krea2 turbo for a week and results are impressive.
I've also completely reworked the entire process to build better-quality prompts for natural language models like Krea. But this will only be available with the July update.

AmvetSB

Eudaimonia and SlideShow 'duration' (not sure if it's a thing in EstimAI 'mode' or not)
Currently there is a static duration setting that can be adjusted during setup, then a prompt at the end of the duration to extend y/n that times out to automatically end (input not required) or selecting not to extend - either of which leads to the grant/deny orgasm, or yes to extend - typing or adjusting from zero to x additional minutes to continue that static value before the option to extend or not pops back up again...
Would it be possible to add some randomness to both the setup duration as well as the end's extended duration??? Perhaps turning the static to a 'maximum' and adding a random option prompting a 'minimum'?
This would allow for committing to window of time but not being in control of the actual duration within that window..
There could also be 'weights' to say closer to the minimum or to the maximum but ultimately still giving up that control to the game. There could be some additional dialogue - or not... Like "I'm done with you, I'm cutting this short" or "Time is up, but I'm going to keep zapping you!"

MalteP

Kairojs, if I use krea2 turbo its seems hard to let him stay at anime or realistic it seems to jump around quite a lot. I used the two loras you mentioned. They also dont always actually uncensor especially if I use a Medium with it. What your current config with krea2?

KairosJS

MalteP I used Krea 2 mostly for realism. Last time, I used both the checkpoint and the NSFW LoRA (SNOFS) linked above. As for the medium, I'm not sure, either cinematic, HDR Photo, or nothing. It was uncensored (too much for my taste, as the model seems to always want to produce nudity), but most of the explicit acts were not well understood.
My tests were with the reworked natural language prompts that aren't in prod yet, which are higher quality.

You may also try using an uncensored LLM as a CLIP model, instead of qwen3vl_4b_fp8_scaled.safetensors.
I didn't notice big changes, but just in case:
Huihui-Qwen3-VL-4B-Instruct-abliterated-comfy
Qwen3-VL-4B-Instruct-heretic-7refusal

For anime, I use Anima (or fine-tunes of it), which is really good and fully uncensored. It generates images in under 6 seconds with the turbo LoRA. They just released an official turbo version of the model, but I haven't tested it yet.

Mountain0fbooks

Thank you for your work!

The sound FX feature would be great it sounds very intresting and would make scenarios more immersive!
The position tracker would also be great, privacy wise for people extra concerned you could offer another open source implementation where the ai runs locally and only the positional data is sent.

Feature Request:
Alltalk2 tts support

Justification

I personally have had issues using local tts because I have an amd card and cpu gen is very slow
uses fastapi like your server.py code
alltalk has support for multiple tts engines (piper, xtts ext)
has an api for standard generation and realtime
Easier user setup compared to pure xtts
Already has kobalt support and a open ai endpoint

Electrodick

A thought... with multi devices. I use a Coyote 3 and an et-312 on stereo. Sometimes though, I only use the Coyote. Is it possible to have a 'switch' that can temporarily turn off a device. When there are 2 devices the stims are more regular across multi channels. When a device isn't used, the stims get a bit sparse on one device.