Depending on which platforms it will support; perhaps a compiled standalone desktop app (like electron) that includes something like a pre-launcher/pre-loader on start up, that checks all required add on are properly installed (If something is not installed the main program wont run.)
I used this method for the Vanilla version; and it was user friendly/hassle free for most part. Ofcourse your application seems way more complex; so i’m not sure how many add on are needed to fully run yer project. Well, goodluck! o7
Looking forward to trying it! Looks very intriguing!
Just FYI in case it wasn’t realized, the video link doesn’t work, and the screenshots are really small / low res 500x241, so you can’t really see any details at this time. Even when you try zooming in. They are just blurry and you can’t read text
Wanted to chime in and echo what @funfunfun mentioned - I’ve been using this since fall of last year on a silicon Mac. And I found the installation process to be super simple. Git clone, download Ollama, download the model, Python venv, install requirements, run app. Six commands copy pasted.
It’s extremely lightweight—the installation package is only 19.3MB. It can be used on a mobile phone; as long as you’re on the same local network, you can scan a QR code to open the Chat page. It can connect to Ollama, enabling 100% privacy protection.
Here’s an early look at my attempt to fix this thing… Most of the obvious problems seem to be solved, but mapping LLM prompts/output to actions on The Handy still needs a lot of work.
It doesn’t seem to be totally psychotic anymore though, and the auto modes work. This is done with LLM tools only, but the quality is already a significant improvement on the reliability front, at least.
You will need a somewhat decent NVIDIA GPU to get the most out of this. I will work on adding hardware accelerated voices for other platforms later.
I won’t be providing support until I have all the features actually where I want them first but I do recommend trying it to see! Ask ChatGPT to guide you through installation if the readme is not enough (paid tier is recommended with thinking on). My main limitation is that I run out of tokens too quickly, and buying extra is too expensive so that slows me down.
You can customize the voice with an audio file. 15 seconds or so is ideal but any voice file will work, the higher quality the better.
Just voice lines, anything else will cause strange behavior though it might give you kinda broken moans and stuff so you can play around with it.
OpenNSFW Sound Pack does not allow usage with AI and the local model is a machine learning model so I cannot include it.
I’ve added a bunch of fixes and customization options since. I am still looking for the best model options, the one that StrokeGPT defaulted to before was not an uncensored model so it worked rather poorly.
If anybody finds a faster/smaller model that works well, let me know. Look for abliterated or uncensored models on Ollama or Hugging Face. You can change the model in settings once you download it to Ollama.
The UI is still crap, but I haven’t really touched it yet other than to add settings, etc. 99% of the work so far is getting a stable codebase.
wow this looks amazing. is there any plan to support TCode standard to be able to drive any device via MultiFunPlayer? or Restim? that would greatly enhance this!
The app is now to the point where I can start working on how it feels, etc. Please give feedback, and share any patterns you design or import in the motion training menu (settings) if you have a chance. Thanks!
StrokeGPT Misa Edition is starting to look like a functional prototype.
Considered a big step-up from Vanilla GPT. (very Experimental in nature).
Will feature the following:
-LITE Dating sim experience.(this is not straight up jerk simulator)
-Multiple varied custom (animated) locations and Activities. (includes lore and trivia)
-Voiceover and custom ambient sound effects for immersion.
-Multiple Custom(location appropriate) NSFW scene’s.
-Some secret unlocks.(depends on User choices)
-Original Vanilla Handy motions have been overhauled to try something new.
-Like Vanilla it will become a stand alone app that can be launched from your desktop and will download required additional files if needed. This Still uses the Handy 1 and firmware 2-3 i believe.
Important noticed: originally the plan was 3D assets but the performance became extremely bad and also was walking on eggshells as a non coder/artist.
Some screenshots below.
I had lots of fun with the Stroke GPT variants so far.
Honestly, the idea to let an AI control your toy works surprisingly well with the Stroke GPT variants, so I’m curious something new popping up :D
Quick question @Zer0 : This is a complete new software, not an addon to Stroke-GPT or Stroke-GPT ReVibed, right?
Hm, i suppose its more like of a new prototype. (thus like Vanilla it will be posted here). It takes past aspects of Stroke-GPT and Vanilla to try something new.
Originally it was gonna introduce Real-time 3d… but the stacking of Atlas Sheets, TTS, Handy traffic, sound/video layers proved extremely crippling during NSFW scenes.
Gave too many headaches for a non-artist/coder like myself. Misa would be a much more expanded/fleshed out idea of what Vanilla was trying to be.
Fair warning: the total file size is no joke… Unlike Vanilla this is gonna be about 7 gb because everything is packed to be stand-alone plug and play for 95% (would still need to download an Model through Ollama).
It cant be compared to peeps that took the idea of stroke-gpt and turned it into something that keeps growing and updating with new features etc.
