Milovana.com

I did some short overview research into the field of AI music generation because I'm convinced that AI should be great for creating stim files.

And look what I found: https://github.com/facebookresearch/audiocraft, released just 5 days ago
Once the code for the training is open sourced (they plan to do it), we can train our own models.

Taken one step further, a tool like https://www.wonder.inc/dynascore/ is imaginable, trained with the best stim files, annotated to specific situations, rhythms and more. We will be able to create stim files on the fly during video production or after. The only thing necessary will be markers for segments and textual descriptions of the wanted stim experience for these segments.


I think it will take a while and some effort to make this viable, but I'm convinced that soon I can say to myself

Sounds amazing. I have some idea of what goes into training, I have no idea how to actually do it. I'm no coder and tend to get lost when trying to understand complicated software things. Don't AI models require large data sets to be any good? Although to us it seems like there are a lot of files, for a learning model its a really small amount.

senorgif2 wrote: Wed Jun 14, 2023 7:58 pm Sounds amazing. I have some idea of what goes into training, I have no idea how to actually do it. I'm no coder and tend to get lost when trying to understand complicated software things. Don't AI models require large data sets to be any good? Although to us it seems like there are a lot of files, for a learning model its a really small amount.

They are normally putting out "xPT model" (x pre-trained model). For our use case, we would need to take source code and build model from scratch by feeding it different kind of files to analyze. It is quite a lot of work.

satn00b wrote: Wed Jun 14, 2023 3:08 pm The only thing necessary will be markers for segments and textual descriptions of the wanted stim experience for these segments.

The sad thing is, this is a lot of work!

edger477 wrote: Wed Jun 14, 2023 8:36 pm [...] from scratch by feeding it different kind of files to analyze. It is quite a lot of work.

I agree.

The thing I'd like to have with an AI thing is biomedical feedback from the cock hero player. This would enable the AI program to customize the experience and edging levels for the individual player. If somebody knows an affordable USB pressure sensor, let me know.

This methodology is really not suited for generation of stimfiles. Stimfiles have very specific characteristics with (in practice) very limited degrees of freedom. These AI's are intended to be used in applications where those characteristics aren't known or their dimensionality is too large to properly encode it.

Restim is (almost) capable of performing dimensionality reduction on stimfiles, if you train the AI on that dimensionality reduced set it's much more efficient. But we probably don't have enough data (and quality stimfiles) to do that.

I have been playing around with pythondancer and restim for automated estim generation from music. It does save time, I scripted a large part of sex cauldron 3 with this tool, but is not a complete solution.

edger477 wrote: Wed Jun 14, 2023 8:36 pm They are normally putting out "xPT model" (x pre-trained model). For our use case, we would need to take source code and build model from scratch by feeding it different kind of files to analyze. It is quite a lot of work.

Indeed, pre-trained models won't work at all. We would need to build a library of annotated stim file parts (morphones) for the AI to learn (i.e. what feels good, what feels like a sharp sting, what is a slow/fast/steady/... ramp-up etc.).

To generate a good model, we need lots of data but since the audio is enough, we probably have a good set of data, even if it is still a relatively small set compared to others (still Dynascore also was trained with only about 1000 pieces from different genres).

gynephilos wrote: Wed Jun 14, 2023 9:10 pm The thing I'd like to have with an AI thing is biomedical feedback from the cock hero player. This would enable the AI program to customize the experience and edging levels for the individual player.

That would be amazing, but will take even longer. There is work on edging detection and some progress on various Discord channels though. Basically the goal would be to prolong the plateau phase nearly indefinitely. But to do this live will be a real challenge.

diglet wrote: Wed Jun 14, 2023 9:34 pm This methodology is really not suited for generation of stimfiles. Stimfiles have very specific characteristics with (in practice) very limited degrees of freedom. These AI's are intended to be used in applications where those characteristics aren't known or their dimensionality is too large to properly encode it.

The strength of AI is exactly to detect correlations that are unknown. Then they can be used to make a new piece of data that preserves this correlation but creates a new sensation (in case of audio file) that works just as well but maybe feels completely different. Or (more likely) feels like a variation of the input.

The goal with annotated training is not to create a stim file from the video contents directly (that's one step further and eliminates a lot of freedom for the creator) but rather to describe the stim experience for parts of the video (e.g. "from 1:20 to 2:24 generate a stim file like ~some anntoated stim file name~ with a steady ramp up, no breaks, no shocks, base frequency 700 Hz, triphase, stroking").

Hi all,

This reminds me of a project I've found on socialstim.

Someone from the community there (fallen_angel_42) already trained an AI Model and provided a estim generator.

The software is called "Stimcomposer":
http://talos.feralhosting.com:23456/


BR
M

Stimcomposer probably doesn't use any AI (at least it isn't mentioned anywhere that is does). But it's also working with samples, though they are much longer than the morphones the AI will be using.

If I understand correctly, Stimcomposer will basically combine a list of annotated, pre-defined stim files while trying to get a combined file that has the (almost) correct length.

Hello,

what about the Project Faraday teases? They should also have "AI generated" stim files.

Bumbum wrote: Sat Jun 17, 2023 7:03 am Hello,

what about the Project Faraday teases? They should also have "AI generated" stim files.

That is just artistic freedom of writing a tease :)

yep, nothing to do with AI as far as I can determine

I followed up on audiocraft and yes, they have released the training code and some examples here: https://github.com/facebookresearch/aud ... RAINING.md and here https://github.com/facebookresearch/aud ... d#training

BUT... it's a mess. Far from being accessible to the general public or even professionals. I looked into it for about an hour now and I'm starting to comprehend how much effort it would be to even start a training (not even considering the work necessary to tag files). It would require loads of programming just to build an environment where you could say something like "here are the files and descriptions, I want a medium sized model, get to work".

If you guys are interested, here's an example of a trainig file with corresponding annotations:
https://github.com/facebookresearch/aud ... ctro_1.mp3
https://github.com/facebookresearch/aud ... tro_1.json

As you can see, this is quite some info. Audiocraft was trained on 20K hours of music. We may not have as much stim mp3, but maybe half of that amount (?). Problem is (apart from the amount of work necessary) that especially the longer sessions would have to be split up because they often encompass multiple "styles" like pleasurable bits, teasing bits, pain bits, edging bits and so on.

A year has passed by and a lot has changed. Today tools exist that enable us to train MusicGen/AudioCraft with our own sound files.



I'm interested to start a project based on these tools. The final result (scope) of the project would be to create tool to easily train MusicGen with stim files, a pre-trained model which can be improved upon by others and a plugin for DaVinci resolve or another video editor which will take text input and a marked area in a audio track and fill it with stim signals generated by the AI.

Now, most of this is "straight forward" with the tools available. However, I don't have a lot of free time and as you may know, AI needs a ton of training data. I simply can't do that alone. That's why I'd like to ask if people are excited by the idea and would like to help cut and classify (tag) estim audio samples. The work would include to cut samples into 30 second sections (best would be of similar content), try them on their stereostim device (we can't support pulse based in the same project) and give them labels like "sharp" "head focus" "throbbing" "gentle" "fast" "triphase" "dual" "orgasmic" "orgasm" or basically anything that categorizes this specific section. The more the marrier. This data then can be used to train the AI.

Anyone aboard?
https://discord.com/channels/7453227717 ... 6609671232