🧬│ai-chat

[glad nebula]

a 50 hour dataset overtrained at 120e

imagine your 2 hour set

it'll overtrain way sooner

[spark agate]

And like you said, technically the lowest point in the graph means nothing right?

[glad nebula]

means nothing yea, just noise

lol

[spark agate]

what are the problems you saw when you went over the 100 epoch range

With the models

[glad nebula]

goes down bc the model learned how to clone silence lols

[spark agate]

Ah that makes sense now

[glad nebula]

the discriminator gets strong and the model starts to have a metallic/robotic sound

past 200e the gen starts to degrade

so the voice starts to glitch badly

for me

weird problems that are absent within the 40-100e range

[spark agate]

but it can all be relative right

Like 200 you are saying it might glitch badly, but not necessarily it will be terrible right? Just worse compared to something around the 100 epoch range

[glad nebula]

it'll be less natural

[spark agate]

got you

And you think og pretrain is the best one to use?

[glad nebula]

you can always experiment, train the full 200e, then compare epoch 40, 50, 60, 70, 80, 90, 100 vs 200

[spark agate]

Yup

[glad nebula]

for me 40-80e always beat 200e

in terms of how natural the model sounds

[spark agate]

oh wow

very interesting lol

[glad nebula]

i have tried datasets of 15 min, 10 min, 5 hours,

and a 50 hour dataset

only one that survived past 100e was the 50 hour one lel

[spark agate]

that’s crazy

[glad nebula]

so yea it's crazy to me when i see ppl saying use smaller datasets too

the training is so much more stable with a bigger dataset

[spark agate]

That’s basically what I was thinking

I just wasn’t too sure

[glad nebula]

if your dataset is good, the model will learn that voice very well

[spark agate]

Yeah my data is all studio sessions

All raw voice

No noise

[glad nebula]

the pretrain voice is not the same as the dataset, obviously it's better to train more hours so you can replace more knowledge of that voice

[spark agate]

ahh

[glad nebula]

but ok lets say more hours don't improve much

would you rather:
A) Small dataset, unstable training, potentially more robotic sound
B) BIg dataset, stable training, less robotic

[spark agate]

yeah B all the way lol

what’s the most realistic model you have trained? Do you happen to have a sample?

[minor blade]

Protools sessions from who?

Juice wrld?

[stark jungle]

bruh, i used veo3 once and £10 was taken from me

[spark agate]

Future

And The Weeknd

[minor blade]

Damn i know bro has only a couple sessions available online

[minor blade]

Weeknd only has 1 session available lmao, and it's from a unreleased Kanye song from 2014

[spark agate]

Yeah he does, I got some other ones a long time ago

[minor blade]

Can you name ones?

[spark agate]

I think that session is lost

[minor blade]

Ye sadly

[spark agate]

Pushin p sessions

I also got lil baby and Travis ones

[minor blade]

Pushin P?????

[spark agate]

But I have to train those

[spark agate]

Yeah lol

[glad nebula]

to my ears rvc never sounds realistic

lol

but making one realistic model is not hard

i think i have one audio of a test model i did back then

iirc this was a spin model? cant remember lol

you just grab some podcast and train it

or studio quality singing

without any denoising or post processing

[spark agate]

thats pretty good wow

accent and everything

[glad nebula]

that is hella old and later ive found some problems with the pretrain

[spark agate]

is rvc the best one for training for now?

[gentle trench]

Just get sora 2

[glad nebula]

imo yes, i got told so-vits-svc artifacts a lot
ddsp-svc requires a big dataset and still i feel rvc quality is superior

[spark agate]

yeah ive tried so vits and it wasnt the best

is training regularly the best way to train atm? or is there like anything else that you think can make models realistic, or is that not possible yet

[glad nebula]

hmm, in a scale from 0 to 10, i'd say rvc is a 8/10 in terms of realism

a good model can really fool a lot of ppl

[spark agate]

yeah very true

do you think there will be anything else in the future

or any updates?

ive been waiting for so long lol

[glad nebula]

from the official rvc team? most likely not
from applio? idk

[spark agate]

damnn

[minor blade]

I confirm THIS

[stark jungle]

Getting that key will be a mission

Oh, I missed its public release it seems.

Azure seems to be giving access so I'll try it out there.

[gentle trench]

good luck

[glad nebula]

yea lol razer managed to confuse me one time

[spark agate]

@glad nebula hey i forgot to ask, about the model i was talking about earlier, i set it to train to 650 but its at 400 rn, and you said anything over 200 around there was pointless, should i just stop trainin git

[chilly lake]

well, this is not true

with a large dataset, say 2 hours every epoch goes thru 10% for random generation

if these two hours vary a lot, then each epoch gets a big varion of those 10% of samples

[glad nebula]

all i can say, there is no diminishing returns for training a big dataset

[chilly lake]

well, that's all I can think of

you have 4x more data, the model does 4x more adjustments per epoch, so may need less training

[glad nebula]

have you put that in practice?

coz i do

none of them but the 50 hour dataset survived 100e

i dont use tiny batch sizes tho

so that must affecting my results too

[chilly lake]

okay, so if you have 30 min and use batch 4 vs 2 hours with batch 16, it should be the same number of steps

that's fine

[glad nebula]

but all of them were surprisingly close, the shorter datasets stopped improving after 40-50e, but the bigger datasets around 60e-80e

[chilly lake]

as long as the number of steps per epoch is the same the only difference is the averaging across the batch

[glad nebula]

just hear the model

[spark agate]

got you

im away so thats why im asking, i left it training all weekend and i monitor it with teamviewer

cuz im not at my pc

[glad nebula]

ur right but i don't think telling people 4 hours is too much, you know that is not true

[chilly lake]

I say it is a waste of time

30 minutes, maybe an hour

[glad nebula]

thats just a lazy excuse to not gather more data to be honest

[chilly lake]

more than that does not bring much extra

[glad nebula]

try using a 30 min dataset in realtime vs a 4 hour one

you'd be surprise which one sounds less robotic

for non realtime is the same story

i just dont see why it can be a waste of time

[chilly lake]

how much of the 4 hour set would be just a duplication?

[spark agate]

i have none

i got 2 hours of all new data

just different tones

and the tones arent too extreme of a change, its just subtle

[glad nebula]

if you train a monotone person reading a book yeah it's completely useless, but if you train expressive speech, is definitely not worthless at all

you see everyone complaining about their egirl model sounding robotic because all of the public ones are 30 mins

is just not enough

same for singers, if you got a lot of data for them, the better the results

it's pointless to rely on the pretrain for singing, you also know how bad and robotic og sounds when singing

[spark agate]

can a model have various tones and be good at them? im not talkimg like drastic changes

just like a bit deep, just a bit and maybe a little baby voice type thing

[glad nebula]

so yea the only time i see it's a waste of time is if the dataset is too monotone

but expressive and diverse dataset? hell naw, train a lot of hours of that

the og pretrain is just a bad base

[spark agate]

wait wym, should i be using another pretrain

[glad nebula]

the dataset should always sound like the same person

[spark agate]

yeah it does

[glad nebula]

if it sometimes sound like a completely different person, delete those clips

[spark agate]

nah i never have any types of audio like that, but theres part of the dataset where the same artist does different tones, like a bit of a baby voice

and a bit of a deeper tone

[glad nebula]

the og is a bad base but we dont have any other better than that lol

[spark agate]

but all sounds the same person

[spark agate]

oh damn, why is it a bad base

[glad nebula]

trained using a monotone dataset

[polar flax]

try excluding parts with roleplayed/falsetto that could be treated as different voice

[glad nebula]

and all of the speakers sound very similar to each other lols

but you can get rid of that base by just training a big dataset

[spark agate]

but still train with og pretrain right

[glad nebula]

yea use og

[spark agate]

is that essentially the reason why my model kinda struggles with naturally doing different tones (i repeat again, not very different, just subtle deeper and baby voice tones)

like my model can handle those but i have to really exagerate them when infering

[glad nebula]

could be both dataset + pretrain

idk honestly

#1408339243948052501 message

[spark agate]

cuz the dataset includes a rlly good variety

[glad nebula]

try mine and see if that happens

[spark agate]

i will have to check out the 2 hour model

this my first time training it

[glad nebula]

if it happens with mine then it could be your dataset

[spark agate]

so i would have to retrain the whole thing right?

[glad nebula]

correct

[spark agate]

alr lol

its been like 2 days

so it might take a while

lmao

but ill try it with a smaller model

[glad nebula]

and dont overtrain pls, you dont need that many epochs

that was a so-vits thing lol

[spark agate]

gotcha

whats basically the benefits of your pretrain?

in a nutshell

[glad nebula]

i noticed less robotic results

[gentle trench]

Can't wait for 3.0

[glad nebula]

(click the link and listen the audios)

[glad nebula]

mine uses both singing and speech and the dataset is a tiny bit bigger than og
24 hours (og) vs 50 mine

[gentle trench]

Just a little

[glad nebula]

still a bit small but hey, it seems to help

at least for me it does improve something

[glad nebula]

have you trained more models with 2.5? how good it is?

[gentle trench]

I've trained a few but I'm not getting good results with some datasets that worked fine with the original legacy core

Even tho I didn't change them

[glad nebula]

interesting, idk why tho

[gentle trench]

Me either

[glad nebula]

the original one used the same dataset as og + singing

but in v2.5 i removed ogs dataset and replaced it with a single speaker one

[gentle trench]

Could be why it changed the results?

[glad nebula]

most likely yea

that single speaker dataset has almost the same hours as og

but im gonna add more speakers in 3.0

that should improve things ig

for me 2.5 works fine tho

[gentle trench]

Hopefully that improves results

[gentle trench]

I mean I train non human sounding stuff like with Venom but that worked fine with original legacy core

Haven't tried with 2.5

I did try Batman tho with 2.5 and never got any good results

What kind?

[royal vessel]

ربات کجاس؟

[vague sail]

I guess that kind of request wasnt welcome

Sorry about that.

[glad nebula]

what kind of bad results? the model sounds more robotic?

[gentle trench]

Somewhat robotic yea, but I feel like it's possible a dataset issue or because his voice is always deep

[glad nebula]

oh yea, og dataset only has deep voiced people, makes sense why your deep voiced model is better with the first legacy core

out of the 109 speakers maybe 2-3 are high pitched? rest is very deep
i can fix this by just adding more guys in the dataset

[gentle trench]

Perfect for 3.0

I'll train batman on 2.0 to see if I can get better results since it has og dataset in it still

[glad nebula]

2.0 no longer has og

[gentle trench]

Oh

[glad nebula]

that one only have the singing dataset

[gentle trench]

Ohhh

Ok

I'll use the original legacy core then

I have each version on my pc

Content vec at least since that's what my ears prefer

[glad nebula]

[gentle trench]

What's wrong with the og pretrain dataset btw, was it denoised?

[glad nebula]

it wasnt able to sing

which is why i added singing to it

the og pretrain can sing because they did some weird hack

ik is weird

[gentle trench]

Must've been ai crack lol

[glad nebula]

lol yea, i tried to make it sing but i gave up, too advanced for me

[strong dune]

Does anyone know if there is a model for the little big planet narrator's voice?

[gentle trench]

not sure, u could make one if there isn't

[strong dune]

I just found out while looking for a model that he claimed multiple times hes uncomfortable with his voice being used with AI.

So nevermind.

[gentle trench]

👍

[full ore]

What's the best game to play on Weights (waiting to get a stronger computer for Replay)

[spark agate]

@glad nebula this is how it turned out

This is epoch 162

[glad nebula]

sounds good to me

[spark agate]

Yeah it sounds pretty nice

You think there’s anything it could realistically improve?

[glad nebula]

hmm idk, maybe you could try testing different batch sizes but that requires training the same dataset multiple times and the difference is usually minimal

[spark agate]

ahh, what exactly does the batch size do?

I’ve never rlly like

Fully understood it

I thought it just had to do with the speed of the training

[glad nebula]

it's the number of the samples the model learns at the time, if you use a batch of 8, the model will learn using 8 files at once until it covers the full dataset

[spark agate]

Ah

[glad nebula]

it can affect the quality, generalization etc

[spark agate]

Yeah I always use 8

[glad nebula]

8 is good yea

[polar flax]

double the batch size, double the vram usage and speed (if there's enough vram allocation)

let's say like when you want to load things on a mule's pack (or a pickup truck) and you don't want to overload it for some reason

[spark agate]

Gotcha

Ty

[brazen ridge]

why they switch to gem things bro

now i cant make free ai models L

[gentle trench]

yes u can dummy, use applio

[dapper drum]

Have anyone try the new Gabox released voc_fv7 beta 1 Mel-Roformer?

[gentle trench]

I can only try it once it's available on the colab for uvr

@stark scarab it's your time to shine

[worthy copper]

is there any chatbot btr than chatgpt?

[edgy meadow]

Hey Yo,
I'm new here, wanna say hi to everyone

[mortal ermine]

Hi

[pliant elm]

bean

[azure bolt]

Hii

[stark scarab]

Me, it will ve available soon on UVR5 UI

[night crane]

i want to create an ai team

anyone interested to work together?

[nocturne frigate]

Hey google, show me this guys jingle bells

[knotty cobalt]

can sum1 send sora invite code

[solar torrent]

https://static.wikia.nocookie.net/blue-archive/images/1/11/Sora_Portrait.png

This girl name is Sora, and she is from mobile game Blue Archive.

[burnt blade]

i am a senior AI engineer with rich experiences in developing project.
In particular, I am focused on Machine Learning (ML),Deep Learning, Natural Language Processing (NLP).
If anyone want to develop AI project, I want to help them sincerely. so if possible , give me dm pls.

[gentle trench]

It's October

[shell pawn]

guys i have question how can i make my Vegeta model not sound screaming while singing

i want him more calm voice

Prince Vegeta (Dragon Ball | ENG Dub) - (RVC V2) 330 Epochs
its this one

but his version of Love sounds calm vegeta but when i try my version he has his more screaming vocals of vegeta

nevermind

i need to put octave lower

[gentle trench]

if u make a model of him u could just remove the screaming from the dataset

[finite stirrup]

need cod for flova please

[gentle trench]

fish?

https://tenor.com/view/hololive-sakamata-chloe-cod-oh-my-cod-vtuber-gif-27689643

[solar torrent]

https://tenor.com/view/pog-gif-14149886028736974766

[static sierra]

Sup

[tired wraith]

anyone alive

[elfin moss]

who got access to sora dm, gonna pay for video

[opaque verge]

msg me for Comet Browser invite

[solar torrent]

Nuh uh.

[haughty hearth]

anybody got a good chinese voice model

[gentle trench]

What even is that

[limber nexus]

a scam

[gentle trench]

Most likely

Scammers and freaks and voice changer abusers oh my

[limber nexus]

ThE GaNgS AlL HeRe

[opaque verge]

perplexity's ai browser?

the invite only one

search it up

[opaque verge]

😭 its legit an invite only browser

search it up

[gentle trench]

https://tenor.com/view/nuh-uh-beocord-no-lol-gif-24435520

[covert lake]

https://www.youtube.com/watch?v=8UWKxJbjriY

[covert lake]

there are thousands of AI Browsers now

Comet was one sponsored by a Discord Quest

now even chatgpt is doing one

[gentle trench]

are these better than regular browsers?

[limber nexus]

too many new browsers, i'll just take my classic, yet still bad Chrome

[gentle trench]

I prefer Opera GX for the ad blocker

even if it steals my data or whatever

[limber nexus]

might have to consider that

[gentle trench]

I like how customizable it is as well

[stark jungle]

Youtube have started to get around my ad blocker. Is OperaGX a solid pick?

[gentle trench]

you still get yt ads sometimes but sometimes not it's kinda random if it skips them

[stark jungle]

Yeah... I have randomness on my end too. I can't figure out whether it is actually youtube developing countermeasures or my extension acting up or catching up to said countermeasures

[burnt garden]

hello

Hey All, I am Final year AI Engineering Student, I need your suggestions for my recent project. if you have 2 min, please DM me , your reviews and suggestions means a lot to me, Waiting for your message ✨.

[stark jungle]

you can ask here https://discord.com/channels/1159260121998827560/1359898289335566570

[burnt blade]

i am a senior AI engineer with rich experiences in developing project.
In particular, I am focused on Machine Learning (ML),Deep Learning, Natural Language Processing (NLP).
If anyone want to develop AI project, I want to help them sincerely. so if possible , give me dm pls.

[chilly lake]

it is a solid spyware

firefox + ublock origin deals with youtube

there are new updates that may get thru, but devs fix them within a week

[stark jungle]

That is honestly my reaction when spammy redirects would send me to an operagx page.

So do I

[modern lantern]

I wonder, would it be difficult to make an AI that can create a chess FEN from a picture? that doesent seem too difficult. Really it would just require the ai to learn to distinguish pieces

[deep minnow]

I'm sure it could, you might even be able to do that just using an existing AI that allows uploaded imagery. IDK if posting commercial systems is ok here, but I can think of a few

[modern lantern]

afaik normal chatgpt or any other ai wont let you do it, though if its possible im guessing someone made one for it already

[deep minnow]

I think if it can see the board and its pretty much any AI its going to know the pieces and correct placement. I'm not a GPT'er so I couldn't tell ya about it there.

[modern lantern]

tested it with chatgpt, it got a bit right, but nowhere near accurate enough.

gemini completely shit itself

[fallen nexus]

are there any free websites where i can train an ai voice

[gentle trench]

applio, u can use it on kaggle

if u need help with it just ask here
https://discord.com/channels/1159260121998827560/1159290139609137264

[fallen nexus]

tysm man

[upper raft]

Hi, could I possibly be guided the right way? I am wondering how some AI music is getting made. The ones that sounds accurate by voice and mood but is in a different genre. The ai music places like Suno and udio does not allow these as well as the lyrics

[haughty cave]

I believe in the future we can connect the AI world into this reality, like legit AI shit just spawning into this world we are living in

[gentle trench]

3D ai generator printer

[peak rivet]

Can someone help me make an ai model ?

[strong dune]

hey does anyone know; any high quality voice models, either free or for sale.
preferably realistic for narration or reading, human-like.
can be male or female, i dont mind either. i really dont know where to look.

[polar flax]

get ublock on firefox, edge or other decent browsers under manifest v2

[sacred bramble]

yo nahh this shit actually insane hahah

some random people believed i was a girl this shi sound realistic 💀 😂

[gentle trench]

https://cdn.discordapp.com/attachments/1385768460348297219/1424641255534624889/image0.gif

[limber nexus]

https://tenor.com/view/chilldog-gif-6253079634849225723

[polar flax]

[sacred bramble]

hi im new to server but i have question

[green mulch]

welcome new member

[sacred bramble]

when i say 1 word with the cvoice changer ON it echos

but

it is off in the NOISE section

is it a way to fix it please guys ?

ah nvm already fixed it

[solar torrent]

For W-Okada the realtime voice changer, better ask in #✨│ai-help, because when you ask about it in chat most people would often ignore or not giving you any solution.

[solar torrent]

There's no paid voice model here; there are free ones in #1175430844685484042.

[high thistle]

Anyone got issue with ChatGPT?

I can't resume any agent chat, and any engine version.

[solar torrent]

For how to train a "model", go to #✨│ai-help or #1192011222023950368 and explain about your goal there.

[solar torrent]

I use Google Gemini.

[solar torrent]

The "AI world" you talking about is when you buy a VR headset, you wear it and you'll see the meta world full of AI people.

[polar flax]

not the metaverse crap

[eternal cipher]

any new updates to okada rvc?

or the rvc in general

or wtv

[vale canyon]

Hi

IAM an AI Engineer

Any one here have a linkedin account

[eternal cipher]

aight then i'll js use the one i currently have ig

[copper cobalt]

AR stuff and such will be very cool as it becomes useful and not weird

I mean Samsung just came out with a much better headset than Apple did a year ago because it actually has incentives and reasons to try it. If you're interested at all, I'm really not but it's way more convincing than anything I've ever seen and I know that the non-metaverse stuff is very good with the quest for people who do like it

I don't think everybody will all switch to that kind of stuff, but recent glasses and stuff like that have been absolutely fantastic

[copper cobalt]

What a weird introduction bruh

[gentle trench]

ew meta, vrchat is the metaverse but not bad

[tight solar]

Hi, everyone.
I'm training a model in RVC 2 using a dataset of 5 hours.
The audio is split into samples ranging from 0 to 18 seconds (a total of 3,811 samples).
Each training step is taking 11 to 12 minutes.
Is this normal?
I'm estimating around 8 days of uninterrupted training to complete 1,000 epochs.
The training is using up 98% of my GPU.

[gentle trench]

so first, u don't need to split the audio into little files anymore as it's done automatically in applio and second use applio not locally so your pc isn't going to explode. Kaggle is the best and easiest non local option

[digital laurel]

why voice changer dont work

who can help me pls

[limber nexus]

Please ask in #✨│ai-help

[digital laurel]

ok

[tight solar]

Thanks a lot for the suggestions, really appreciate it!

I'm currently using Mangio-RVC v23.7.0 locally with an RTX 3060 12GB, so my idea was to take full advantage of my own hardware and avoid cloud time limits.

I was mainly wondering if the current training speed I'm getting (around 11–12 minutes per step) is expected, or if I might have some misconfigurations causing this slowdown.

Again, I appreciate the help and all the input!

[gentle trench]

oh lord

mangio is like

ancient tech

I am so glad I helped you

training speed will depend on your hardware and how long the dataset is along with batch size relative to dataset length

but with applio it's pretty fast

[tight solar]

Mangio really is an old warhorse, but I’m giving it a chance.
I’ll try to reduce the dataset size to speed up the epochs.
Thank you for your time and the tips, I’ll consider Applio as an option for my toolkit.

[gentle trench]

good luck with your journey, and againI am glad to have helped ^^

[dawn temple]

Hi! Before I post anything, I just wanted to ask: is it okay to request a Sora 2 invite code here? I don’t want to break any rules. Thanks!

[gentle trench]

It is but most likely nobody will answer about it since there are some people that join here and go "sora code" like cave people

But I believe there is an actual server to get a sora invite code or you could ask friends that may have one

Good luck cookie monster

@hardy spear why'd you dm me? You could just message me here

[covert lake]

you won't break the rules, but it will just be 99.9% probably that you won't get any codes here

[dawn temple]

okey thanks anyway 👍

[arctic haven]

Hello!

a question, theres a model here that looks like the link is death

DidDoes anyone happen to have any backups?

Himemori Luna

[vestal lotus]

Imagine a future where AI provides everyone with a guaranteed income and meets all basic needs. In this world, people only work if they want to, for creativity, passion, or personal fulfillment. How would you feel about living in such a society

[solar torrent]

You say it like you hate everything you hate huh.

[solar torrent]

What do you mean the model is dead? I think there's one in #1175430844685484042.

[arctic haven]

Yes, but it cannot be downloaded

[polar flax]

sorry that the huggingface repo link has gone private, but you could still search it in weights site

[arctic haven]

There is a weights site?

Sorry i'm kindda new in this

[solar torrent]

The Mangio RVC is quite outdated. There's Applio which is a better RVC fork.

[polar flax]

try this #🔍│find-models message

[tight solar]

Hey everyone!
I'm currently training an RVC voice model for the Male Nord voice from Skyrim, but my GPU will need to run for almost 23 hours nonstop to finish it.
I was wondering if anyone happens to have the Male Old Grumpy voice model from Skyrim (RVC .pth file) and could share it.
I’d really appreciate it — it would save me a lot of training time. Thanks in advance!

[solar torrent]

You can train another voice model but it will take some time to finish training for sure. Another thing, for RVC, you can go to #✨│ai-help and explain about your progress on which you stuck the most there.

[tight solar]

Yeah, I’m thinking about retiring Mangio. I’ll do some tests with Applio before making a final call lol.

[arctic haven]

Look like I found it, but still I can't download it

ah but there is another one

[polar flax]

in the weights site?

[arctic haven]

yep

a wait

I found the button

jhahaahhaa

nvm

thanks!

[gentle trench]

What

[polar flax]

no one willing to give it but if they do, they'd be more likely a scammer

[grim locust]

hello everyone im using Applio and Chatterbox for creating ai voice. can anyone recommend me a good AI tool for generating video that is open source?

[stark jungle]

Well let me tell you something about that.
There would certainly be people who choose to do more despite having basic survival taken care of. These would be those that have ambition to be at a better standing in life or want to contribute to the world or simply want to enjoy their hobbies at the professional level.
On the other hand there will be those that choose to be content with the basics and decide to not contribute to the productiveness of society or more realistically there will be limited opportunities available to them.

In a world where AI is prevalent, a world where survival needs are taken care of for people would lead to a population growth and we would see way more people than opportunities to do anything because AI is doing everything. The quality of life for people may hit an all time low.
My personal opinion is that in order for AI and people to live together, there needs to be less people so that opportunity congestion and more importantly, population congestion and resource contention is mitigated and quality of life for the individual person is improved.

It's either have less people or colonise other planets.

[glossy valley]

hi

is there any way to get midjourney?

trial or something

[worn condor]

yo can anyyone help me the setup i changer isnt change my voice

[spark agate]

i accidently quit my rvc while it was finishing training and i didnt get to download the index

can i still get it without training a new model or

[storm bane]

holy yap

[graceful lake]

hello

[stark jungle]

That was a summary too

[graceful lake]

is there someone currently building something related to ai?

and is it feasible to talk about such topic in this chat?

[storm bane]

remind me to never argue w u

[stark jungle]

Naturally you can talk about ai in the ai chat

[graceful lake]

okay

[stark jungle]

Why, are you looking to collaborate?

[graceful lake]

yes something like that.

finding someone who can share a project with me or two

[stark jungle]

Well I do have a project that wouldn't mind some more researchers or testers.

[graceful lake]

you made a gambling site i see

is that your project?

[stark jungle]

Nah its a Generative ai platform

[graceful lake]

haha yes i joined the discord to know more

[stark jungle]

@barren kettle damn, didn't know ai could do all that, it sounds production ready too

[graceful lake]

it still haven't started i guess.

[stark jungle]

It's in closed beta but keys are given out periodically so don't worry about it, your time will come

[graceful lake]

oh nice.

you read mangas?

[stark jungle]

There isn't many I'm reading right now but I do

[graceful lake]

basically my plan is to create a free manga reader website where users and creators earn crypto tokens by mining in-browser, with user-shared resources powering the site and supporting manga uploads, all without hosting costs or traditional payments.

cause nowadays many manga creators don't earn money. sites can't give money cause they use it in resources for servers and storage and no extra fund. and noone reads manga by paying.

ads are the only way and most free sites are shitlike.

[stark jungle]

@barren kettle is this really all ai, its hella good, are the lyrics really written by ai??

[stark jungle]

New crypto tokens are dead now that people have caught on to the pump and dump scams. If I were you I'd focus on how AI could be used in manga

[graceful lake]

no i just using token system. my site runs on readers computer and uses their computing power to power my site and indexing. and they get tokens to skip ads for a time being.. or use tokens to cheer their fav creator.

[stark jungle]

Hmm I see where you are going with that. Good luck with that anyway, I've abandoned the crypto space for now. Unless there is an opportunity for me to combine web3 and ai in the future.

[graceful lake]

it's just an idea. i got yesterday.

AI and web3 needs lot's of money to spend behind to make it proper so was thinking from starting with small thing.

don't forget to give me beta access to your ai

[stark jungle]

Yes, I envy the startups that can raise millions and then dissolve because of "technical limitations"

[graceful lake]

yes

want to try lots of things but money is the only limitation.

[stark jungle]

C'est la vie, as long as keep on assessing the viability of the ideas you come up with, you can make one stick even without money by having a plan for small scale usage, then work your way up. It is the only way for those with no connections or millions in the bank.

[graceful lake]

true

you french i guess

[stark jungle]

I am not but I regularly use quotes from many languages

[graceful lake]

oh i see.. no worries see you later

[neon marsh]

hi guys

new here,any interesting topics

[neon marsh]

i think its a great feeling wish it comes true

[uncut verge]

Hey guys!

[neon marsh]

hi

[icy basin]

Hi

[tall light]

Meows cutely

[lament crest]

@graceful lake
Your plan is really promising. it combines decentralized resource sharing, tokenomics, and community incentives in a way that could completely change how manga distribution works.
here’s a high-tech spin on your idea
Use IPFS/Arweave for manga storage so there’s no server cost, and let users’ browsers cache & serve pages to each other, kinda like browser-based proof-of-contribution instead of heavy mining. Creators earn tokens whenever people read or share their manga, plus you could do optional NFTs for rare chapters. Throw in some AI for recommendations & auto-tagging, and maybe a DAO-style system so the community decides spotlight content and rewards.
Basically: readers power the site, creators get paid, and the whole thing scales without servers or ads.

[gentle trench]

Nfts are cringe, I'll copy and paste the image

[main nest]

people playing meaningless buzzword-bingo again? 😁

[gentle trench]

https://tenor.com/view/pita-a-horse-is-here-gif-3838550382478913993

[undone tartan]

how do i use the ai voices? lowkey new to this ngl-

[gentle trench]

are u trying to use them on a voice changer or make them sing?

[undone tartan]

voice changer, like i just want to make my own ai song, js wanna test it out

[gentle trench]

fair, I can help u out in https://discord.com/channels/1159260121998827560/1159290139609137264

[undone tartan]

Okay bet!!

[burnt blade]

hello

[brittle nebula]

someone send me carti athena raw vocals

[tight zenith]

Hello, I want to create my first model on my phone, but I don't know how.

[solar torrent]

I don't know who "Hina" is where they made W-Okada for Colab/Kaggle; the Hina I know is Sorasaki Hina and Hina Kagiyama.

[solar torrent]

Voice model or the acapella? If you're looking for acapella, this server might not be the one you could find.

[solar torrent]

Hi.

[split basin]

HOW IS THERE BARELEY ANY KING OF THE HILL VOICE MODELS!???

[solar torrent]

See #1175430844685484042.

[ripe sable]

Quick question does anyone have issues while using pollinations on python? It will always skip every single parameter I set for the generation and just make a simple 768x768 img

[brittle flare]

Hi there

[polar oracle]

do i need to have a headphone and a proper mic

to use a voice changer

[solar torrent]

For Stable Diffusion or related features, you can ask in #✨│ai-help.

[solar torrent]

Yes, both microphone and headphone/speaker are needed for W-Okada. Virtual Audio Cable is also needed. For W-Okada the realtime voice changer, go to #✨│ai-help or #1192011222023950368 and explain about your issue or progress there.

[gentle trench]

https://tenor.com/view/oh-moderators-gif-25571015

[obtuse frost]

GUYS

[trim oasis]

antonio xavier freeman jr.

[gentle trench]

[pastel tundra]

https://www.youtube.com/shorts/6wMRngKgS2c does anyone know what kind of program this youtuber use for text to speech voice ai?

[midnight sentinel]

I 2nd this question

[violet totem]

Does anyone know how to create ai influencer

[hollow beacon]

Hey everyone! 👋
Quick question, is it okay to share my own AI-generated videos that are uploaded on YouTube, or should I only post direct content (images/videos) here in the server?
Just want to make sure I’m following the rules correctly. 😊

[mortal ermine]

[solar torrent]

More like an AI VTuber.

[plush wasp]

Hello, I'm looking for a recommendation. I'd like to be able to swap the voice of a person in an ASMR .mp4 video or .mp3 audio file with another person's voice. I'm on Win10, 5060Ti 16GB. So far I've locally generated images and animated images.
I've searched the ⁠🎧│voice-models channel and I haven't found a model of the person. So I'm guessing I'd have to be able to train one too. There is plenty of sample audio, thankfully.

[wide dome]

is there a updated better guide for creating ai models

i have 30mins+ of studio vocals and i want the best model

[solar torrent]

It's in https://docs.aihub.gg/rvc/resources/training/ For more information, you can ask in #✨│ai-help.

[swift quartz]

Pipeline is not initialized.
[Voice Changer] Waiting generate pipeline...

How to solve the problem

[red storm]

https://x.com/FearAvion/status/1982140050042216689

[floral pagoda]

I've never used kaggle before, is there a way to start the realtime rvc server from the dashboard or something like that instead of having to redo the whole tutorial?

[gentle trench]

Is your computer not strong enough for wokada :(

[nocturne maple]

wsg

[molten dune]

the voice is delayed by like 30 secs

[rancid birch]

my voice is lagging how can i fix it

[pine flame]

where do we get roles?

[cloud island]

can someone help me with the new voice changer? i cant download it

[cloud island]

can you help me, i cant download the nvdia voice changer

[spice rose]

where do you usually find game character's voice dataset?

[violet totem]

Can anyone tell me how I can create ai model

[spice rose]

which model to be exact?

[violet totem]

Like influencer

Real looking

[spice rose]

like video and image, or else?

[violet totem]

Yes

[spice rose]

if that's the case, you can try to install comfyui or reforge

[violet totem]

Are they free

[spice rose]

yes, completely free, provided that you have good gpu

[violet totem]

Ok cool thankyou will check it out

[violet totem]

Are these human like?

[minor meadow]

you need to download models also

[spice rose]

yeah, based on the model you choose, and lora

[violet totem]

Ahh I see

[spice rose]

there is a dedicated discord server specialize on this stuff, it calls Stable Diffusion

you can check that out

[minor meadow]

I guess wan is a bit better now I tried both's quantised versions BTW

[violet totem]

Is there videos on how to do it

[minor meadow]

yah all over youtube you can find

[violet totem]

What do u type in to utube

[spice rose]

yeah, it also has large community support

easy to find lora

[minor meadow]

yeah Now i'm taking a break lol it's like every week the chats update for something new

T_T

[minor meadow]

search like comfyui install or how to use text to inage in comfyui stuffs

[spice rose]

hold on, do you have a gpu?

[minor meadow]

she will install lol

[violet totem]

Thankyou guys I appreciate it

[deep sluice]

yo I'm JJ, been working on something called Render a Discord based AI that adapts tone and responses based on context. it's in early testing rn, if anyones into bot development or AI behavior layers. I'd love some feedback or people to test with.

[topaz jackal]

Cool

Something similar to alfred ai?

[deep sluice]

sorta yeah, but Renders build more around adaptive behavior and context like if you talk to it casual or seriously, it adjusts tone and response style. I'm expirementing with emotional layers instead of just commands

[topaz jackal]

Alright. Will it have humor?

Cuz why not

[inland root]

Wsp

[deep sluice]

yea thats the goal to make it feel like it actually gets you not just replies humors a big part of that

[plush wasp]

Anybody got some advice?

[tidal onyx]

Is there any chat model similar to grok ai's level to run on Koboldcpp?

[still sigil]

in terms of intelligence or conversation

[tidal onyx]

intelligence

i heard mistral nemo 12b iis good for roleplays or convos, but thats not what im looking for

[still sigil]

give this a shot: https://huggingface.co/mistralai/Magistral-Small-2509-GGUF

[lime raven]

i have a (maybe) interesting idea

what if someone made a visualizer for rvc that uses a playhead marker to show what specific vowels/syllables it used to convert an audio for example?

i feel like itd constantly erratically jitter around

[deep sluice]

so kinda how you learned the ABC''s growing up?

[lime raven]

what

wat do you mean by that

[deep sluice]

when you learn to talk or learn the vowels its constantly repetead and it shows specfic vowels kinda the way you originally learned it

[lime raven]

youre somewhat in the right direction?

maybe i should visualize this

ill try something

[deep sluice]

bet

[lime raven]

gimme a second

[lime raven]

something like this

[deep sluice]

ohhh

interesting

[lime raven]

i really wanna see a real visualization of that

but i dont think anyone would take their time to do that lol

[deep sluice]

it would b cool if someone did tbh

[lime raven]

ye

[lime raven]

but i feel like itd look very similiar to my example of it

i guess albeit more erratic

[deep sluice]

yea, it could also b good for music producing so if it did come to b for music and prolly content

[lime raven]

music producing?

[deep sluice]

yea i could see them with artist using it to correct stuff or find the right sound

[lime raven]

hm maybe?

[deep sluice]

prolly beats

[lime raven]

do you mean like drum voice models

[deep sluice]

i think so

[lime raven]

them using this this hypothetical rvc "original input sample finder" (idk wat to call it)

[lime raven]

oh

i guess i can see it

[edgy light]

Hi there

How's Applo doing

[gaunt finch]

guys what is vonovox ??

[topaz jackal]

W

[hasty geode]

Yo

[rigid steeple]

i just downloaded a voice from the weights website and the download speed is VERY SLOW.

like everytime i download a voice, the download speed is at megabytes but it gradually decreases to ketabytes (whatever its called) until its 0

[hollow ginkgo]

what pitch should i put for the e girl voice model?

that sounds the best

[gentle trench]

Ew

[rapid ice]

want to troll in games huh

try harvest or fcpe

tune 12-13 depends on your tone

[hollow ginkgo]

alr

[hollow ginkgo]

yeah

i do gng 😭😭😭

[edgy light]

Scammer

[limber nexus]

No advertising

[limber nexus]

aint no way, a literal gpu, Hello!

i'll have you know i only have IGPU, not a dedicated one

how are you?

good to hear, im pretty good

nah, i only have my sassy AI, haven't talked to her in a while

i would gift you image perms but Voya screwed me over earlier, so i don't have enough

voya is the bot here in charge of the server currency, you can buy different perms and roles with the server currency

we can but in #✦│chat not here

[normal heron]

i need help

[rocky robin]

Aye do these voice changers rlly work for

Fr*

[rapid ice]

yes

u need a good pc

and configure it

[lament plinth]

Hello friends, I'm looking for an AI that can read files for game programming.

I installed LmStudio on my computer, but I don't know which AI to use. My computer specs are as follows:

Monster T5 23.1
RTX 4060 laptop
32GB DD4 3200MHz RAM
i5-13500HX CPU 8-core Vcore 16

[rancid iris]

BLUEFACE BABY

YEA AIGHT

[fickle siren]

?

what just happend?

[spice rose]

how can i use tensorboard on kaggle?

[solar torrent]

That one question sounded awkward, even if the annoying "E-girl and catfishing" prohibitions have been removed from help guidelines.

[tired fog]

Hey everyone,

I’m currently studying how companies and startups run design partnerships and would love your take 🙏

Any brief notes on the questions below would mean a lot:

-When you look for a design partner, what must be true about them (profile, stack, urgency, data access)? How do you gauge real intent vs. tire-kicking before committing time? Any signals you trust?Where/how do you normally find design partner candidates?
-What value exchange works best (discounts/credits, roadmap influence, support SLAs, exclusivity windows)?

-What does a smooth, end-to-end design partnership look like in your experience?

-Where does this process slow down (security, scope, etc.)?

Huge thanks in advance! Even a handful of bullet points is gold!

[solar torrent]

Just get some helps from people you love or can help you.

[solar torrent]

W-Okadas do work, but work best with a powerful dedicated GPU like NVIDIA GeForce RTX.

[elder willow]

guys why cant my voice work on RVC models but it works on beatrice ones

[solar torrent]

Vonovox is an AI realtime voice changer, a complete alternative one to W-Okada. It simply as that.

[solar torrent]

For W-Okada realtime voice changer, better go to #✨│ai-help or #1192011222023950368 and explain about your issues there.

Many people have been asking for the same thing in #🧬│ai-chat when there are help channels for that, and most of which did pretty much the same mistakes when joining server and start talking.

Still not where you would yap.

[polar oracle]

ty mr weights by namari

[gentle trench]

I still think nobody should help those people

[rapid ice]

guys can i use the vcc in write mode or smth?

[weak token]

anyone got sora 2 codes?

[little frost]

hi, is there a french version of this serv ?

[stark jungle]

pretty sure the app is out now

[weak token]

ios only tho

[bleak pivot]

guys, I have 13 hours of clean audio from AiMER's voice, I want to train a RVC model for her, how many epochs should I run?

I was aimning at 1000 but I read it could be overkill somehow?

[ripe spade]

applio takes zip files as datasets?
here's the structure:
Model.zip └Model └file1.wav └file2.wav

[ripe spade]

There's no way to guess, use tensorboard to look for overtraining

[bleak pivot]

soo what's tensorboard and how do I use it?

[polar flax]

how about test the model which is better at <100 epochs or 1000?

[bleak pivot]

yeah I plan to left training overnight and see what I get in the morning and go from there, but just wanted to make sure

[polar flax]

I suppose it could take more than a day on a 5090 unless aiming for less than 100 epochs

[gloomy grotto]

yo can anyone help me with smt

[bleak pivot]

It could take a month for all I care

[gloomy grotto]

im tryna create a minecraft mod where u can do somethig like "???whats the recipe for a strength 2 potion?" and then the ??? calls the ai and it gives a response. I wanna use chatgpt api key to connect my mod to chatgpt servers so the response generates from chatgpt and shows client side on my mod

[stark jungle]

should come out on play store soon, currently im testing it out on azure, pricing is stupendous good compared to veo

[ripe spade]

https://docs.aihub.gg/rvc/resources/training/#tensorboard

[polar flax]

how about test several checkpoints on the cloud or another pc while having ongoing training

[bleak pivot]

thank you!!

[bleak pivot]

I'm super noob and only using the RVC webui at the moment, so not sure how I'd do t hat

[polar flax]

keep yapping your excuses

[bleak pivot]

if you have a noobfriendly tutorial I'd super give it a go

[polar flax]

the guide is there #🧬│ai-chat message and I can't afford to do step by step babysitting

[stark jungle]

get chatgpt to make it noob friendly for you

[bleak pivot]

tensorboard is okay, I meant the several checkpoints on the cloud

[polar flax]

you just need the pth file for particular epochs (and better if with index file)

[bleak pivot]

I'm saving every 50 epochs so I'll have that in a moment or two

[polar flax]

that's too much, even every 1-10 epochs is fine

[bleak pivot]

thanks! I'll adjust

[polar flax]

im not sure if the guide still recommends that

[sage sierra]

where did the egirl models go

[weak token]

How good is it?

[gentle trench]

https://tenor.com/view/all-roads-lead-to-rome-clock-ticking-white-rabbit-alice-in-wonderland-gif-11005866172848930328

[sage sierra]

yes

wheres E-women

[gentle trench]

no shoo

https://tenor.com/view/spray-bottle-cat-spray-bottle-spray-bottle-meme-loop-gif-25594440

[steep sinew]

...

[stark jungle]

Well I hate to burst your bubble but they've clamped down on the copyrighted stuff and it will probably only become more restricted so no more detective pikachu. NSFW is also currently disabled so no goon for you

[weak token]

Nah, i'm planning to make different kinds of slop :)

[stark jungle]

Kinda hard to say it is better than Veo 3 because Veo still does some things better like lighting and atmosphere making it fit for movie making however Sora 2 has better reasoning when creating some scenes.

Perhaps it all comes down to mastering the prompts for each model...

[stark jungle]

Still.. it is cheaper than veo 3 for me at the moment so the tests shall continue.

[weak token]

How much are both? Don't really know the prices

[stark jungle]

Hah, through vertex veo costs an arm and a leg

Well, the version I was using anyway, there is a cheaper and faster alternative but I ain't tried that.

Sora 2 is probably cheaper than veo 3 fast version

The price will change when it becomes fully accessible though

[vestal lotus]

The Architect Cycle

The Architect Cycle is a theory proposing that highly advanced civilizations eventually reach the limits of their own creativity. At that point, they can no longer generate novel ideas themselves — either because they have explored all possibilities available to them, or because further creation feels futile or inefficient. To continue evolving, these civilizations seed or observe other worlds, allowing new species to develop unique minds shaped by their planet’s resources and environments. Once these civilizations reach their creative peak, the Architects harvest their innovations — technologies, philosophies, and cultural developments — before resetting or moving on to the next world. In this way, the cycle sustains progress across the universe, with each lower-dimensional civilization contributing what the Architects themselves can no longer produce.

[stark jungle]

For veo, honestly depends coz Google being a bit of a dick about it by offering so many different ai plans and subscriptions.

[vestal lotus]

i made this theory above , is it tuff?

[gentle trench]

https://cdn.discordapp.com/attachments/1365750911070371972/1428134698495901797/Soapspeechbubble.gif

[stark jungle]

Vertex once charged me 10 dollars for a veo 3 ultra prompt....
As for sora 2, Azure foundry offering it for like 80 cents per 8 second video.

[vestal lotus]

so is a new future side hustle gonna be a prompt writer?

[stark jungle]

It's already a thing

[polar flax]

it's just a bubble, I heard it is getting devalued, otherwise the job listings might be more likely an utter scam

[main nest]

<@&1159293140440723499> 👆

thx

[dry torrent]

Hi

[burnt harbor]

Hi

[mental dagger]

someone knows a good ai voicechanger in realtime

[solar torrent]

Yes, some members and helpers here know about W-Okada, where one of the helpers is me.

[mental dagger]

is it free?

[solar torrent]

For W-Okada realtime voice changer, better move the topic in #✨│ai-help and explain about your operating system and PC GPU there.

W-Okada is a free and open source program.

[lime raven]

random weird question that might not make sense but:

rvc is essentially just a audio to audio synthesizer (albeit finetuned for human speech) right

thats why it can accept (some) sounds like saw waves, instruments, drums, etc aswell

[solar torrent]

You could train a voice model with any audio, even drums and noises, as long as it works as expected.

[lime raven]

so my analogy is correct right?

[chilly lake]

not exactly

[lime raven]

o

[chilly lake]

the model has two parts - one takes input phonemes, pitch + trained speaker and predicts a spectrogram

another parts takes the spectrogram and turns it into audiousing a bunch of filters

you can take an input of drums or piano that are in the speech range and the model would basically make an audio of a person beatboxing

violin - not so much

[lime raven]

what if i make it so that i play individual pitches of the instruments

and make that the input audio for the model to train off of

[chilly lake]

you can take an audio of someone playing a piano, bass guitar or an electric guitar, train on that, then try to infer speech and it would sound like someone trying to emulate speech with a guitar

[lime raven]

wouldnt there be less "noise" if it the input audio was of regularly playing the instrument and instead a chromatic scale so rvc could recognize the audio better?

[chilly lake]

it is less about playing individual notes

thing is, the contentvec extractor is not exactly trained on music

[light needle]

I’m interested in training a pretrained anybody know where I should start

[queen kernel]

Collect a clean dataset of 250 house with 100s of speakers and then, get a 5090 to train it. 😊

[light needle]

Then what

[whole onyx]

what voice changer should i use

[gentle trench]

What gpu do u have?

[whole onyx]

intel n100 and intel uhd graphics

[gentle trench]

What is a n100

O

Intel can't do AI very good at all, AMD even struggles so you're cooked for doing it locally, u could do it on a browser but that comes with weird issues and idk how to set that up at all

[whole onyx]

aw

[gentle trench]

Ye you might have to buy a desktop PC instead of using a laptop

[solar torrent]

Not always.

[polar flax]

anything works but the pitch extraction methods may struggle on polyphonic sounds like choirs, piano chords, etc (in this case you can try crepe)

[noble basalt]

a mobile gaming company filed some patents on ai with really impressive performance / effeciency ehancment on a bunch of differnt things.
One system watches a language model while it’s generating text and blocks it from looping or babbling, so you get the answer in fewer steps and spend less power to get it. Another one removes duplicate data across memory and disk so the machine stops moving the same bytes over and over. A third one monitors neural network training in real time and reallocates GPU, memory, and bandwidth on the fly so you aren’t burning energy on stalled parts of the pipeline. At the OS level there’s a scheduler that kills redundant work between processes and frees RAM. And at the hardware level there’s a processor architecture that actively routes work and power only to blocks that are doing useful computation, reuses charge instead of just dumping it as heat, and cuts wasted switching.
If this holds in practice, the impact is blunt: lower cost per token at inference, lower cost per checkpoint during training, cooler data centers, and higher throughput per watt. In other words, scaling AI by squeezing out redundancy and reusing energy instead of just buying more GPUs.
copy pasted stats from the press release :

1. Computer System for Optimizing Text Generation

Addresses inefficiencies in large-language-model inference by detecting and preventing redundant token patterns.
Key Results:
• 40-50% reduction in output redundancy
• 25-35% decrease in GPU/CPU cycles per useful token
• 20-30% faster generation latency
• 18-28% lower power consumption

2. Computer System for Data Storage Optimization

Uses locality-sensitive hashing to eliminate redundant data across cache and storage systems.
Key Results:
• 30-50% reduction in physical storage requirements
• 25-40% higher cache hit rates
• 20-35% lower I/O bandwidth usage
• 15-30% faster data access

3. Computer System for Efficient Neural Network Training

Enables real-time monitoring and automatic resource optimization for ML workloads.
Key Results:
• 20-30% faster training time
• 25-40% lower GPU energy use
• 15-25% less memory bandwidth utilization
• 15-30% improvement in model accuracy

4. Computer System for Automated Resource Management

Provides OS-level control of physical computing resources through redundancy detection.
Key Results:
• 20-40% less RAM usage
• 15-35% lower CPU power draw
• 25-45% lower I/O bandwidth
• 30-50% higher system throughput

5. Processor Architecture for Improved Computational Efficiency

Introduces specialized circuits for real-time optimization of physical resource allocation.
Key Results:
• 20-35% lower energy use
• 15-40% fewer execution cycles
• 18-28% less thermal output
• 45-65% increase in operations-per-watt efficiency

Integrated Technology Stack:
Together, these innovations deliver compounding efficiency gains across the technology ecosystem:
Application Layer: AI text generation optimization
Framework Layer: Neural network training acceleration
Operating System Layer: Resource and storage optimization
Hardware Layer: Processor-level efficiency design
This end-to-end approach addresses inefficiencies at every computational layer, advancing a holistic vision of sustainable, high-performance computing.

[gentle trench]

the fuck?

[cobalt imp]

guys can u tell me what is batch size and kl and mel

i dont know those things

[noble basalt]

in a good or bad way ?

[gentle trench]

more in a I'm confused becasuse wall of text

[noble basalt]

yea thats mostly just because i copy pasted the stats

i thought it was worth tho to show the diff

[gentle trench]

fair

[polar flax]

you dont need to

and it seems nothing more than some babbling about a chinese AI infrastructure with everything in-house

[noble basalt]

alot of it (if not all) is new tech

which is why there patenting it

although i can see how it dosent really say what the actual technological improvments are if you plug the equations into an ai and prompt it correctly it will be able to show what the actual improvements are quite easily

from new algorithms to new chip designs its a pretty broad spread and a very vauge (word bable) press release

[queen kernel]

Nothing just train it.

[polar flax]

I didn't see anything meaningful on those % improvements, the article should show comparison charts against other LLMs (e.g. gemini 2.5 pro, claude sonnet, grok, etc) for vram usage, coding, reasoning, etc.

[noble basalt]

yea its quite far from being an actual product (any of them) the press release was just talking about the pataents they had filed very vaugly but when they get through aproval then they become publicaly accesable and hopfully some properly benchmarked results come out

[polar flax]

that's why we should take it with several grain of salt

[deep sluice]

First time building something like this would anyone care to give me some feedback

[ripe spade]

Is there any way to auto backup the lowest epoch?
Like, if the epoch 28 is the lowest, keep that file until a new lower point is reached. Ex epoch 57 is the new lowest, now it will delete e28. Does it make sense what im trying to ask?

[chilly lake]

dont use overtraining detector

just save every 10th epoch

[gentle trench]

I save every 5

[covert lake]

that pc setup is not good for any local ai task, either get a better pc or use cloud (remote good pc), you can check the ai hub docs and ask for help #1192011222023950368

[smoky bolt]

ts sound to ai

[muted locust]

Job

[gentle trench]

you can't say that it's scary

[ripe spade]

not what I asked

I was trying to say that since we look at some graphics to determine where our model was at it's peak, and Applio knows which was the best epoch so far in the training, why doesn't in just store that epoch for us? Instead of saving every X amount of epochs

or just doing both actually.

[chilly lake]

"since we look at some graphics to determine where our model was at it's peak, " we don't

[ripe spade]

so tensorboard is useless?

[chilly lake]

You can use it to see whether the training had suddendly explode, but it does not give you a feedback for how good the model can infer test files

you can only hear it yourself

all the training metrics are applicable only to the training dataset

[ripe spade]

I see...

So it's not just "lowest epoch is better"

[chilly lake]

not really, much sooner than that

[lilac coral]

yo

i put all the right setting the output on the voice changer is input and the input on discord is output

please helppp

[limber nexus]

please ask in #✨│ai-help and provide more info

[lilac coral]

done thx

[gentle trench]

what

[gentle trench]

learn to do it yourself, it's free and easy

all u really need is uvr and i guess izotope

[steel sigil]

I've been looking through gwen3 coding for websites, but it just can't get the asthetics right and just gets into a loop and runs out of context. And i'm not paying 100 bucks for claude. Anyone know what i can do?

and also i was gonna exploit the claude free trial but it's not there rip

[mighty fern]

gg

dad

[craggy leaf]

CHAT

what r the rules

[limber nexus]

#📜│rules

damn, bro, idk who it was but tell a mod

[polar flax]

https://tenor.com/view/horny-jail-dogs-gif-14221969916105575779

[gentle trench]

literally just look at the name of the person dummy

[limber nexus]

https://tenor.com/view/no-gif-no-gif-perms-gif-27679658

[gentle trench]

damn

[limber nexus]

you gotta get to level 5 by chatting, im not giving you gif perms

cuz im still recovering my server currency

well go chat or gamble

[polar flax]

[polar flax]

[limber nexus]

in #🤖│bots you gamble, i told you, im not giving you media perms, theres a reason you don't already have them

[spark agate]

hey @chilly lake sorry to bother, i wanted to ask if you knew if having autotune in a dataset mess the model up?

even if the quality is good

[polar flax]

how about try it first, also still the dataset shouldnt be too short

[hollow laurel]

hwt

[maiden sedge]

Are there any better female voices?

models

[chilly lake]

if you want your output to sound autotuned af

[pallid thorn]

Hi, I’m a full-stack developer based in Singapore with deep expertise in AI automation and AI agent development. I’m passionate about building scalable systems and collaborating with forward-thinking teams to solve real-world problems through innovative software solutions. Open to technical partnerships and impactful project opportunities.

[spice rose]

for those who have finetune rvc, tts with the voice dataset with high pitch, complex intonation like paimon or march 7th, how can you do that?

[limber nexus]

[mystic sail]

hi, I am new here, can I get a gist of what is this all about

[limber nexus]

Um wdym

[gentle trench]

You're gonna get executed

[sour oak]

No

Not good

[limber nexus]

NO ADVERTISEMENT

[gentle trench]

those things that do that are not real ppl there's no way

[molten root]

Hey everyone is anyone currently building a start up? Would love to hear what youre building

[gentle trench]

is that the robux symbol?

[limber nexus]

i think so..

[main nest]

It's actually a Benzene ring with a circle, part of Unicode version 5.0 onward , U+23E3,
The circle inside it relates to the concept of resonance in benzene. I don't know enough about organic chemistry to explain what that actually means, but that's what the character is meant to represent, and why it's in the unicode.

Why anybody would try to steal another person's benzene is a mystery to me, though.

[gentle trench]

fuck is a benzene?

[main nest]

It's a chemical compound, one of the elemental constituents in petroleum and petrochemicals in general.
You could've just googled that, though.

[gentle trench]

What does this have to do with ai lmao

[main nest]

I don't know, you were wondering about it!!!

[gentle trench]

I was wondering about it bc it was randomly put into this channel out of nowhere

prob me being slow but it looks like it has nothing to do with ai

[main nest]

Hey fam, if you can afford that server and a marketing team you can afford paying somebody with the technical expertise to answer all your questions, instead of spamming it across multiple channels on random discord servers. Worm Regards.

[mossy wadi]

hey thanks for the reply, acctualy its not my server just a donation from a friend that i can use, also am a technical person, got to this server because all people here love ai and i love ai please if u can help me

[solar torrent]

Artificial intelligence.

[tame reef]

i cant launch my ai voice changer what do i do

[covert lake]

Please elaborate in #1192011222023950368

[covert lake]

https://x.com/sama/status/1984025727763935585

Sam Altman (@sama)

GPT-6 will be renamed GPT-6-7, you're welcome

X · Oct 30, 2025, 10:32 PM

[gentle trench]

13th reason/j

[cobalt badge]

this before actual music cover apps on mobile that are not weights is crazy.

[covert lake]

huh

[covert lake]

what? If you're talking about running RVC Locally on phone instead of cloud, even tho it's possible, it's not suggested since phones aren't good enough to run ai task that good

[covert lake]

https://www.reddit.com/r/LocalLLaMA/comments/1okz8qz/pewdiepie_dropped_a_video_about_running_local_ai/

[foggy snow]

who knows how to create AI models, please contact me

[gentle trench]

there are many diff kinds

[covert lake]

elaborate in #1192011222023950368

[north breach]

whos french here

I need ai hub france

[limber nexus]

i am living the life ive dreamed about, life isn't about money

[gentle trench]

https://tenor.com/view/all-roads-lead-to-rome-clock-ticking-white-rabbit-alice-in-wonderland-gif-11005866172848930328

[lilac rapids]

Bros be doing everything except AI hahahaa

How u guys doing today

[solar torrent]

You mean this? Because I don't know, their Discord server link has long been expired.

[gentle trench]

wait it's real?

[limber nexus]

i didn't know there was a AI Hub france

[solar torrent]

I've long heard about the existance of AI Hub France or AI Hub French since 2023, but never joined one. I've never seen anyone here talking about it since 2023, aside from Automaze bot spamming its bot command with the same phrase in several text channels.

[jolly perch]

question but what would be good settings for my rtx 3050 and when i play roblox it lags when using it prob because my cpu is a ryzen 3 pro 2200g

[solar torrent]

For W-Okada realtime voice changer, better ask in #✨│ai-help or #1192011222023950368.

[solar torrent]

You can call anyone a tourist, but this is an Discord server about artificial technology, not X/Twitter.

[gentle trench]

Bad, go away you foul animal

Catfishing is illegal

[umbral osprey]

give me some prompt about Cyberpunk Video

[finite hazel]

comet vs atlas
wdyt ?

[floral pelican]

One message removed from a suspended account.

[drifting panther]

Looking for an AI tool the turn logos into stinger animations.

I saw one of the obnoxious "powerful websites you should know" videos online that featured a tool that did it without template and seemed to actually be using some form of model generating animations. Can't find the video now

[drifting panther]

If anyone else is looking forward in the future, I think it was magic animator but I'm not for sure. Will do more testing.

[noble halo]

what tools are people using for ai images that don't cost anything?

[hidden river]

how do i fix the voice cutting out while i talk?

[limber nexus]

Local Generation

[limber nexus]

Please ask in #✨│ai-help and provide more info

[obtuse bobcat]

checking in case:
Im trying to run on CPU an STT, whisper medium int 8, with an API integration. running in a docker container.
Ive tried 3 diff github premade images, not working. Ill try a build myself etc. but in case some1 did something similar holla at me please.

[noble halo]

What’s that? How do I get local tools tho

[limber nexus]

well most people here (at least the ones that post images here) use comfyui, idk much on where you download it or how to use it, but im sure you can ask

[noble halo]

Darn okay. I just wish someone could walk me through this. Ik it may not seem complicated but as someone who’s tried like 50 times it just doesn’t make sense 😭

[humble dove]

guys can someone turn my voice into an ai model

and face

[lapis edge]

Can someone please help me with So-vits?

i sent a message in the help channel but it didnt get replied to

[spring spruce]

When people talk about the "Original Pretrain" Which one do they mean, TITAN?

[gentle trench]

absolutely not

they mean the one used by default

when no pretrain is selected that one is used

[covert lake]

you can ask for help in #1192011222023950368

Cloud = running the generation on a remote server, ofcourse there is a paid tier because it costs to do this

Local = running the generation on your own good hardware, so you need a good pc and to use only open source ai tools

[covert lake]

the default one

[covert lake]

use #1192011222023950368 for help

We don't allow selling models since a while, and sending discord invites would get you automodded

[north plaza]

https://garticphone.com/ru/?c=02826feba0

[tawny pumice]

someone help me

it doesnt work

[floral pelican]

One message removed from a suspended account.

[polar flax]

no one would care, please go to https://discord.com/channels/1159260121998827560/1159290139609137264 and consider doing this:

!howtoask

[pine acorn]

❓ How to Ask for Help

✅ Before You Ask!

1. Check Docs & Guides: Your answer may already be in the AI Hub Docs or the https://discord.com/channels/1159260121998827560/1159513888199540817 channel.
2. Search the https://discord.com/channels/1159260121998827560/1192011222023950368 : Look for existing posts that solve your issue. Do not invade someone else's post.

📝 How to ask?

Tell your:

• Full GPU Name: (e.g., NVIDIA RTX 4060 8gb vram desktop)
• Operating System: (e.g., Windows 11)
• Detailed Description: What were you trying to do and what went wrong?
• Tutorial Used: Link to the guide you were following.
• Screenshot: A picture of the full error message.

🚫 Prohibited Topics (We Will NOT Help With These)

To maintain a legal, safe & ethical community, we will NOT provide help for:

• ANY illegal activities.
• NSFW/Porn.
  Requests for these topics may be ignored, not helped and result in moderation action.

<:matsuripray:1159685390156967936> Community Expectations

• Be Polite & Patient: Our helpers are volunteers. You may ping the Helpers role once.
• English Only: Please keep all conversations in English.
• Don't Ask To Ask.

[wispy tide]

yo

[sturdy summit]

Is there any new updates for MMVCServerSIO

[gentle trench]

?

[sturdy summit]

I've download MMVCServerSIO Voice changer in the past

Now idk where the link is

And is there any new updates?

[gentle trench]

there's much better stuff now, what gpu do u have?

[sturdy summit]

3070

It has delay and idk sometimes I cuts off like it doesn't record everything but I think it's because of my web cam

[gentle trench]

alr we should prob move this to https://discord.com/channels/1159260121998827560/1159290139609137264 but u could download vonovox it's much better than ur current outdated software

[sturdy summit]

Alr

[spark agate]

Will rvc 3 ever happen

[polar flax]

not even before GTA 6 or half life 3

[spark agate]

But you think it will actually happen?

Or is this it for rvc

[polar flax]

why should I think so?

[spark agate]

huh

[polar flax]

https://tenor.com/view/guy-arguing-guy-talking-to-wall-talking-brick-wall-gif-18667615

[spark agate]

Took you that long to find that?

Im asking a question idk what you on abt

[polar flax]

https://tenor.com/view/chicago-fire-stella-kidd-what-confused-huh-gif-2808668890218747779

[spark agate]

im speaking about rvc3 with someone who cant even comprehend what im saying

🥀

oh my lord why are you still typing

take your time g

Get that sentence out

[polar flax]

https://tenor.com/view/stop-talking-be-quiet-stfu-hit-by-a-radio-gif-17151302

[spark agate]

🥀🥀

[spark agate]

time to get off discord for the day

[polar flax]

I think you're nothing but a random who doesnt know anything about #🔊│ai-development

[spark agate]

Rainbow name

Opinion rejected

💔✌️

[limber nexus]

https://tenor.com/view/what-huh-wat-wut-gif-13031409

[polar flax]

https://tenor.com/view/bait-cat-is-this-bait-cat-is-this-ragebait-gif-10902073105296387701

[spark agate]

holy cope

LMAO

[spark agate]

thank you 🙏

[limber nexus]

Sorry, not the way i meant it

[polar flax]

I dont get why he talked like that

[limber nexus]

idk, entitled people i guess

[spark agate]

didn’t mean it sarcastically btw

[spark agate]

victim mentality 🥀🥀

[limber nexus]

i know, that's why i said what i said

[spark agate]

🔥🔥

Probably are

Yeah

Delete that message like a good boy

😂😂😂

[limber nexus]

No

[spark agate]

Im not talking to you

[limber nexus]

okay, good

sorry

[polar flax]

@dawn temple you could even demote my mod role and strike me cuz I feel like a loser here

[spark agate]

plz just stop, im trying to do stuff without getting pinged 50 times

just let it go plz🙏

gotta drag everything smh

[polar flax]

thats my line, except I dont mind not being an escapist from here

[spark agate]

Yea yeah type shit

[polar flax]

I do

[spark agate]

yeah

[polar flax]

not sure if ur a weirdo here, or me, or even both

[spark agate]

yeah

[polar flax]

so both true

[spark agate]

yeah

[dawn temple]

What's going on?

[tender arch]

yo

[nova scarab]

Idk if this is against the rules to talk about but is jailbreaking gpt 5 with prompts a thing

[gentle trench]

🤷‍♀️

[mental vapor]

Обезьяна

[compact lava]

wheres the download link bro

for the appp

[dry falcon]

bro what app?
the voice changer?

[rare meteor]

can someone help me solve glitching/cutting

[sleek meadow]

Anyone down to rate and give feedback? DM me, want honest opinions on how I can improve my look

[covert lake]

@rare meteor @dry falcon @compact lava
**This is a General AI Server, please check #1402790586028789830 and elaborate your help requests in #1192011222023950368 **, we can't really help in this channel and can't understand which app you're talking about

[rare meteor]

Alr ty

[remote zealot]

ai

why this server is dead

[jade nexus]

I’m working on an early-stage AI platform in the B2B sales intelligence space, I’m looking for a co-founder to help build and scale the product from the ground up.

[gentle trench]

Yes this is an ai server 😮

[coral narwhal]

U already know why im here

[gentle trench]

u wanna see men kissing obviously

[elder willow]

Sounds interesting! What kind of co-founder are you looking for tech, biz, or a bit of both?

[slow mountain]

Is there any free ai tools i can use to make those Minecraft stories

[covert lake]

for help use #1192011222023950368 or #✨│ai-help

[reef totem]

cant you just make your own sounds for exaple tturn a clip of your friend to pth