Bro wtf this is literally what I'm working on right now Edit: Just went through the video, here are my thoughts - Is this not open sourced? - You can use an embeddings model alongside a vector database to store long term memory for her. - For "remembering the current conversation", instead of passing the entire conversation back to the prompt as "memory", you actually just pass the conversation to the LLM in a separate prompt asking it to summarize the entire convo, then pass the result back into the personality prompt as context. Combining this with texts pulled from your vector DB can allow the AI to recall long-term memory as she makes more and more conversations. - Using an embeddings model can also help with categorizing the incoming stream of chat, for example you can check which of the chat messages are question, reaction or request etc... - You can also control the vtuber model by categorizing the AI reply into different states of emotion, like happy or sad etc...
@@gadiel4351 By using Semantic embeddings. The idea is that the embedding model converts text into a vectors that hold the semantic/meaning of that text. Then when you have a new question you can just queries for conversation pieces with the most similarity and pass it to the prompt as context.
I fking tried to make a vtuber model using stable diffusion but I couldn't figure out how because it didn't draw in layers. Then I fell behind in classes and had to put the project on hold. Super excited to watch this vid.
Doing God's work. Great editing. The fireworks are beautiful. 8:51 True Enlightenment Thank you so much for the tutorials! It's really possible! Thank you for the inspiring step-by-step guide.
dude u always have the best/most random themes for ur videos and i love it i hope ull continue to create content like this. i dont know how hard it is to create things like this but respect to you anyways❤❤❤
wow great video, didnt expect this at 1:00 am lol very interesting to see you share part of how you did this, I learned a lot and nothing at the same time because you are at 67/100 and I am at 12/100 on computer skills and understanding!!!!
Literally getting into coding because this is the train of thought I've been dreaming about. Nice to see you cracking on with these projects and really powering through, not only experimenting but sharing your methodology. Subbed
loved the bit where you had one instance generate the training data for the description expander. Took me a while to work out that's what you were doing there.
My friend, you are amazing and a potential programmer. With the number of videos you share, it is worth at least one UA-cam channel with a few hundred thousand subscribers, even up to millions. 😵
This is awesome and this give me tremedouns amount of joys hope and envy, because man you did everthing i want to do with AI, since i have learned about "neuro-sama" : Creating from (almost) A-Z an AI Using the same said AI for streaming and integrating it into game made withe Renp'y, wich totaly blew my mind As rigth when i was discovering your chanel only a few a week ago, i was wondering if this was even possible but you litteraly blew up my expection and proved me that was possible and just for that a BIG thanks, while i don't even know if this comment is going to get read by the time i publish this comment, just thanks from france anyways Also just to give a small idea of what i mean by creating an AI from A-Z, its quite litteraly, start with nothing from start, not even proprer knowlege for coding and AI nor LLM, just having one initial objectif : Creating something that could at least speak french, and maybe one days speak with other AI while playing some game... Wich migth be ambitious at the time considering that when neuro-sama got out, i had 0 knowlege of how she was done or i was suppose to do that, wich lead med to wait for month, until i get chat gpt-4 wich basicly helped me to create my AI until now... And does it work...? Actually yes, while initialy the AI was eating my CPU instead of my GPU, i still managed to fix that issue... Homever since i am stuborn and dumb, i did not succeed or know how to use local LLM such as Vicuana, wich lead me to build my very own LLM ! Wich is currently composed of only few 1000 words/sentences for the moment... And the first sentance that i got from my AI was "Je est bien" wich translate itself by "I and good", wich considering the fact that its has pretty much no ressources or was trained only a few 5000 time (wich is ridicule considering that chat-gpt4 recomend, at least a training of around 50 000 for 1k sentences...) i have to says that its did pretty good... ok i'm migth be to optimist and very stuborn for willing to put myself througth this, but i don't care And now, knowing that you can do that with local LLM, i will use in my advandage to hoppefully be able to reach the 10k sentences and having my very own little streaming AI, and while a tutorial video dedicated to how you've donne everthing precisely, step by step, explaing everthing would have been appreciable, i will still take what i can take, and one more time thanks !
Ngl, good sh1t man, i hope you bring back talking to random in Tokyo servers though, that's one of the reason i found you when you're on double digit subs, it's INSANE how fast you grew 28k now, 100k soon
Hilarous, informative, inspirational and no code for fuck sake gimme code to look at !!! XD Also rainy chan needs to become a recurring thing. What would happen if you put a bunch of them together ? Even better, ask subscribers to build their own and have all of them in a chatroom talking to each other And I give you the Weeb award, you're clearly winning at weebing. If you don't know what the f to do with your life, you're doing something really awesome here :D Have a good one daddy dearest
I am confused how this is all free? He mentions using cloud computing and Google cloud hosting which maybe has some free use but certainly after a limit they charge and I imagine making this you'd soon hit your limit of requests or storage etc
I want to build this one, but I do not understand any bit of it. Any advice where can I start? I mean from someone who do not know any programming at all. Any guide or tutorial is really much appreciated.
ROFL I want my own one of these if for no other reason than to hear these absurd rants “Cucumbertarianism” “Due to their resemblance to fetuses” LMAO WAT On a side note, I’m genuinely impressed it knew to pronounce Wei Wang properly instead of like “Wei WaYNG” Soooo refreshing to hear ❤
I was trying to make my own vTuber using UE4/5, but sadly didn't work out >.< The cost of implementing the monster was way too much for me, heh. Might go the 2D route instead and just make it a standard vTuber, not a 3D vTuber that is daytrading stocks and talking about stock market news and the economy, heh.
zheesh never had to pause a video so much in my life! Thanks for the explanation though. So did you use the edge plugin function for the voice in the end or did you use the azure text to speech API direct? Does the edge one need an account btw?
11:01 Who cares? Worrying about the inevitable is just pointless and ultimately stupid this can only be depressing for people who worry about the sun setting.
Honest question cause im lost, (i have only small programming knowledge so this might sound super dumb) im trying to create this project so im also running Vicuna on google colabs and im trying to do the thing where he runs it on colabs and connects it to his local project by making it a web service??? Please explain this part also the colab notebook doesn't stay on forever soo.. how do i fix that part? Note: i know it's not much of a clear question but in very lost and i really want to complete this project
Hi i have a question for you or anyone else here who can help? Ive got an ai already set up locally and im trying to turn it into a chat bot for social media, ive got the api working but it doesnt seem to use the characters i made instead just giving generic responses
Is there any text to speech ai than sound human (the one u use in the waifu apex video speak japannes and I want one who speak English or other language)
Maybe use Linux until the complete ROCm stack come to Windows. I have the whole thing working locally on a 7900xt the voice and the AI. I used lama.cpp and koblodcpp. oobabooga use lama-cpp-python so it should work too.
Did you use something more than you provided in the Renpy game where you let the AI decide to talk and what to talk? Cause I can't quite get it to work.. is there some hidden things you didnt tell? :(
Bro by the end of the year everyone will be able to create their own AI Vtuber waifu lmao and I'm all for it
Truly, the best timeline
@@mikec1222 no
Bro wtf this is literally what I'm working on right now
Edit: Just went through the video, here are my thoughts
- Is this not open sourced?
- You can use an embeddings model alongside a vector database to store long term memory for her.
- For "remembering the current conversation", instead of passing the entire conversation back to the prompt as "memory", you actually just pass the conversation to the LLM in a separate prompt asking it to summarize the entire convo, then pass the result back into the personality prompt as context. Combining this with texts pulled from your vector DB can allow the AI to recall long-term memory as she makes more and more conversations.
- Using an embeddings model can also help with categorizing the incoming stream of chat, for example you can check which of the chat messages are question, reaction or request etc...
- You can also control the vtuber model by categorizing the AI reply into different states of emotion, like happy or sad etc...
Excuse you!? How can you use vector as long term memory?! How did it works?
@@gadiel4351 By using Semantic embeddings. The idea is that the embedding model converts text into a vectors that hold the semantic/meaning of that text. Then when you have a new question you can just queries for conversation pieces with the most similarity and pass it to the prompt as context.
I fking tried to make a vtuber model using stable diffusion but I couldn't figure out how because it didn't draw in layers. Then I fell behind in classes and had to put the project on hold. Super excited to watch this vid.
👀I think I need information.... help.. with something like this 🤔
that sounds like a great idea! Do I understand and complete what you are saying? No I am not because I understand nothing being said!!! words
I admit this is a pretty cool idea very well executed. Kudos
Doing God's work. Great editing. The fireworks are beautiful. 8:51 True Enlightenment
Thank you so much for the tutorials! It's really possible! Thank you for the inspiring step-by-step guide.
Rainy sure sounds like a barrel of fun. Would gladly watch her stream in a miniplayer while I do mundane office work.
dude u always have the best/most random themes for ur videos and i love it i hope ull continue to create content like this. i dont know how hard it is to create things like this but respect to you anyways❤❤❤
wow great video, didnt expect this at 1:00 am lol very interesting to see you share part of how you did this, I learned a lot and nothing at the same time because you are at 67/100 and I am at 12/100 on computer skills and understanding!!!!
Glad you passed your driving exam and the CompTIA CySA! GG 🎉
Never stop uploading you'll go big with that hard work bro i love the vids !!!!
Hope you keep with the vídeos, love the format that you explain. Have a good day :)
Literally getting into coding because this is the train of thought I've been dreaming about. Nice to see you cracking on with these projects and really powering through, not only experimenting but sharing your methodology.
Subbed
YESSSSSSSSSSSSS❤❤❤❤❤ THANK YOU ALL THE OTHER PEOPLE DONT TEACHE YOU NOTHING 😢... BUT YOU MY MAN YOU DID IT THANK YOU SO MUCH I FINALLY FIX MY WIFU 😂❤❤
loved the bit where you had one instance generate the training data for the description expander. Took me a while to work out that's what you were doing there.
This is really interesting stuff. Congrats on getting your certification.
it's SIWeeb again dropping some absolute banger that's completely free.
what a god damn hero
Keep up the videos friend! I'd love to pick your brain when you have the time. GL with the life stuff.
My friend, you are amazing and a potential programmer. With the number of videos you share, it is worth at least one UA-cam channel with a few hundred thousand subscribers, even up to millions. 😵
Me continuing watching this guy creating one after another God work.
This is awesome and this give me tremedouns amount of joys hope and envy, because man you did everthing i want to do with AI, since i have learned about "neuro-sama" :
Creating from (almost) A-Z an AI
Using the same said AI for streaming and integrating it into game made withe Renp'y, wich totaly blew my mind
As rigth when i was discovering your chanel only a few a week ago, i was wondering if this was even possible but you litteraly blew up my expection and proved me that was possible and just for that a BIG thanks, while i don't even know if this comment is going to get read by the time i publish this comment, just thanks from france anyways
Also just to give a small idea of what i mean by creating an AI from A-Z, its quite litteraly, start with nothing from start, not even proprer knowlege for coding and AI nor LLM, just having one initial objectif :
Creating something that could at least speak french, and maybe one days speak with other AI while playing some game...
Wich migth be ambitious at the time considering that when neuro-sama got out, i had 0 knowlege of how she was done or i was suppose to do that, wich lead med to wait for month, until i get chat gpt-4 wich basicly helped me to create my AI until now...
And does it work...?
Actually yes, while initialy the AI was eating my CPU instead of my GPU, i still managed to fix that issue...
Homever since i am stuborn and dumb, i did not succeed or know how to use local LLM such as Vicuana, wich lead me to build my very own LLM !
Wich is currently composed of only few 1000 words/sentences for the moment...
And the first sentance that i got from my AI was "Je est bien" wich translate itself by "I and good", wich considering the fact that its has pretty much no ressources or was trained only a few 5000 time (wich is ridicule considering that chat-gpt4 recomend, at least a training of around 50 000 for 1k sentences...) i have to says that its did pretty good... ok i'm migth be to optimist and very stuborn for willing to put myself througth this, but i don't care
And now, knowing that you can do that with local LLM, i will use in my advandage to hoppefully be able to reach the 10k sentences and having my very own little streaming AI, and while a tutorial video dedicated to how you've donne everthing precisely, step by step, explaing everthing would have been appreciable, i will still take what i can take, and one more time thanks !
just wanted to say this video is amazing and RAINY is a top tier waifu for real, not lying, no sarcasm, not sarcastic, 0% fake, all real, and real😊
Dame the animation is so cool 😅
Ngl, good sh1t man, i hope you bring back talking to random in Tokyo servers though, that's one of the reason i found you when you're on double digit subs, it's INSANE how fast you grew 28k now, 100k soon
I think I'm falling for you, weeb...
Hilarous, informative, inspirational and no code for fuck sake gimme code to look at !!! XD
Also rainy chan needs to become a recurring thing. What would happen if you put a bunch of them together ?
Even better, ask subscribers to build their own and have all of them in a chatroom talking to each other
And I give you the Weeb award, you're clearly winning at weebing. If you don't know what the f to do with your life, you're doing something really awesome here :D
Have a good one daddy dearest
You should do a tutorial on how to do something like this! Also really good video
Is there any chance for a step-by-step tutorial for people that don't know much about programming, and would like to try to create own AI Vtubers?
1. Learn coding
@@minnscat 2. Learn how to make an aAI
@@minnscat Still would need help, some tutorial, knowledge of python language isn't enough.
I guess no, at least now
@@FinGeek4now How do you debug your codes?
this is really cool ngl this can really become something big you should dif keep streaming with rainy
my brother i beg of you to create some kind of tutorial about how to go about making something like this my life may literally depend on it
Would absolutely love a educational video step by step by step on how to do this making your own GPT-text and making your own GPT-art and vtube-AI
Thank you, brother. I will make her REAL! JFC I'm so lonely.
im tooooo damn excited
yep... another banger
Gotta move the livestream into the cloud too (server that has obs installed)
I am confused how this is all free? He mentions using cloud computing and Google cloud hosting which maybe has some free use but certainly after a limit they charge and I imagine making this you'd soon hit your limit of requests or storage etc
DON"T YOU DARE PUT HER DOWN!!!
Truly amazing...and it's also cool...
Neuro-sama but with no soul
This is just awesome.
You have inspired me to learn coding... I shall learn and then create my own AI... 🤩🤩🤩🤩
I want to build this one, but I do not understand any bit of it. Any advice where can I start? I mean from someone who do not know any programming at all. Any guide or tutorial is really much appreciated.
i was waiting for you to use bark Ai but that goes against what you want because its more realistic
ROFL I want my own one of these if for no other reason than to hear these absurd rants
“Cucumbertarianism”
“Due to their resemblance to fetuses” LMAO WAT
On a side note, I’m genuinely impressed it knew to pronounce Wei Wang properly instead of like “Wei WaYNG”
Soooo refreshing to hear ❤
Would love to see the full code for this for inspiration for my own AI Vtuber i am working on
this will be amazing
Amazing work man
yes we need more stream, also hoping rainy -chan will able to draw herself😆
That's pretty impressive!
3 words : damn thats EPIIICCC
I was trying to make my own vTuber using UE4/5, but sadly didn't work out >.< The cost of implementing the monster was way too much for me, heh. Might go the 2D route instead and just make it a standard vTuber, not a 3D vTuber that is daytrading stocks and talking about stock market news and the economy, heh.
awesome video! will you share the prompts that you used for the model? I'm really curious about that
zheesh never had to pause a video so much in my life! Thanks for the explanation though. So did you use the edge plugin function for the voice in the end or did you use the azure text to speech API direct? Does the edge one need an account btw?
Can't Wait!
Rainy Supremacy
i rather support AI vtubers than real people
the Vtuber model is art of a vocaloid I believe lol how ironic 😭
Tuttorial please
the goat
The greatest mystery that I cannot solve is text to speech. Pretty much all of it is hosted on the cloud via a paid service.
11:01 Who cares? Worrying about the inevitable is just pointless and ultimately stupid this can only be depressing for people who worry about the sun setting.
Sir teach us. It would be much appreciated.
Bro the video was amazing and so was Rainy-chan! I just wish you'd stop using ai that uses stolen art
She's a gem lmao 😂
Can you give a longer step by step tutorial on doing something like this?
im shook
Real women cause depression, AI waifu cures depression. It became a strange world.
you're amazing.
4:20 btw how does paperspace work with this?
Ohhh so amazing.
LET'S FUCKING GOOOOOO
YES
I've tried using vicuna model, but it is generating prompts by itself, how did you stop that from happening ?
Also it forgets itself after sometime 😂
Honest question cause im lost, (i have only small programming knowledge so this might sound super dumb) im trying to create this project so im also running Vicuna on google colabs and im trying to do the thing where he runs it on colabs and connects it to his local project by making it a web service??? Please explain this part also the colab notebook doesn't stay on forever soo.. how do i fix that part?
Note: i know it's not much of a clear question but in very lost and i really want to complete this project
>personality
>utaha
10/10
Cool.
If you are the case then i say yes 😁
Hi i have a question for you or anyone else here who can help? Ive got an ai already set up locally and im trying to turn it into a chat bot for social media, ive got the api working but it doesnt seem to use the characters i made instead just giving generic responses
Awesome
How to generate subtitle like that? Did i miss the explanation?
Yo yo yo it's getting out of hand
impressive
Actually Virtual UA-camr was created!
where’s the tutorial
Fun fact : kiwis provides much more potassium than banana
Detailed tutorial please! :D
maaan that is awesome work u made this AI is submissive and breedable
Is there any text to speech ai than sound human (the one u use in the waifu apex video speak japannes and I want one who speak English or other language)
Awesome
Will you do another live stream with rainy Chan
Bro do anyone here know how can i get Vicuna language model?
CODE BRO CODE PLS SOURCE
This guy 👍👍
honestly shes better than neuro sama
Cool!
Any idea on how to remove latency?
I was trying to do the same, but the course is using AMD GPU and I can't make oogabooga works on windows with AMD gpu, Why did I choose AMD why?
Maybe use Linux until the complete ROCm stack come to Windows. I have the whole thing working locally on a 7900xt the voice and the AI. I used lama.cpp and koblodcpp. oobabooga use lama-cpp-python so it should work too.
does it work to have some sort of Linux simulator? because my pc is running on windows. I will look into it. Thank you@@leucome
1:13 me 😎
just 1 single question : something like api from chatgpt is pay to use ?
Did you use something more than you provided in the Renpy game where you let the AI decide to talk and what to talk? Cause I can't quite get it to work.. is there some hidden things you didnt tell? :(
Neuro and Rainy collab stream when?
So where's the live VTuber? give us a link. :)
WWOWW :OOO