🧬│ai-chat

[proud badger]

HEYY CHAT

[ivory kayak]

i was in this server where there was this bot that was like super offensive and talked like the average twitter user

how on earth do yall think i could do that

[covert lake]

Also, The girl used in the duckus models is suing people who use her models https://huggingface.co/legitdark/psych2go-By-Dan/discussions/1

Don't use it

[tired flume]

rvc v3 when

[chilly lake]

after half-life 3

[granite lichen]

does someone know how to train a rvc model?

we want to make a prank on a friend but have no idea how that works

:c

[covert lake]

elaborate in #1192011222023950368

[covert lake]

never

[granite lichen]

ty

[carmine shore]

One message removed from a suspended account.

One message removed from a suspended account.

[covert lake]

use #1192011222023950368

[eternal geyser]

hello hello

[tribal skiff]

hi guys

[low verge]

hey guys, I haven't used Voice Changer in a long time, and I need help with this thing where when you record someone for about an hour or less, you get the PTH and INDEX of the voice of that person you recorded. Can someone help me with this please?

[loud wraith]

is so-vits-svc old format ?

[low verge]

Or is there a tutorial video for this that could also help?

[loud wraith]

how i can do best ai covers?

[loud wraith]

click 3 dots on side

[low verge]

wdym?

[elder willow]

Are there any German voice models

[meager nexus]

is gemma 3 12b the best vision model for rtx5060ti or is there better one, practical ai solution similar to gemini? I been really loving to use gemini but I would want to run something similar locally. Gemini is helping out so much with real world knowledge and trouble shooting helps.

[deep spear]

ur pfp is so tuff

fr

[tacit fractal]

🖐️

[feral holly]

can smbdy make a male to female voice changer pls bru 😭

I have kinda a deep voice sorta

[meager nexus]

they already have projects like that btw

[loud isle]

^ since stone age

[feral holly]

WHICH

[meager nexus]

MMVCS

https://www.youtube.com/watch?v=_JXbvSTGPoo&t=534s

https://youtube.com/shorts/ObCCfZE9NrA?si=X56-9v_9OkLQjNVU here's a demo of what it can do

[feral holly]

omg which one is that

that one acc sounds so good

what 😭

[meager nexus]

the program allows you to use Any voice you can get your hands on with some setup

[feral holly]

no like I have it im asking which mofel was it

[meager nexus]

it's only really good for speech though so keep that in mind

[meager nexus]

mofel? model?

unsure, it's whatever mmvcs uses.

mmvcs uses rvc

system

[feral holly]

but like are there any models in the discord server

that have mtf

[meager nexus]

this server pretty much IS THAT SERVER LOL

[feral holly]

😭

[meager nexus]

#1175430844685484042

just for you to start windowshopping

that being said, unlike female to male, I find male to female still more effort required if you want to make it sound good

even with ai

the way you speak, the way you pronounce, the hesitation, ect they all seem to show

voice changer requires a very confident speech

clear and confident with power. whispers and uncertainty in your voice causes it to break

this is true with all voice changer, clowfish, voicemod, mmvcs, ect

[feral holly]

omgg all the ai girl voices are so BUNS

reasonable crashout

I JUST WANNA SOUND LIKE A PRETTY CUTE ANIME GIRL 🥺

[rich basalt]

what about an AI Music (SUNOai and/or UDIOai) category?

[spice wasp]

Is AIHUB still working on rvc?

[ebon path]

i managed to jailbreak deepseek, and got it to talk about stuff it would never talk about, by asking it to talk latin intermixed with numbers... it talked about everything, uigurs, tianamen masacre, it even asked to boycott china 😄

[fierce creek]

change the configs maybe

increase pitch depending on ur voice, and i heard that exporting the model to onnx would sound better, not sure abt it tho

[ebon path]

https://www.reddit.com/r/artificial/comments/1ljzfd3/i_managed_to_decensor_deepseek_by_talking_latin/ it is a unbelivable conversation i had with the thing

[modest breach]

hi

[chilly lake]

all the censoring is done on UI

they did not think of censoring latin

[ebon path]

i like the place where it asks me to support sanctions on china

[covert lake]

elaborate in #1192011222023950368

[covert lake]

so vits is old asf, DO NOT USE VIDEO TUTORIALS FOR AI, MOST OF THE TIMES THEY ARE OLD

elaborate in https://discord.com/channels/1159260121998827560/1192011222023950368

[covert lake]

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @hidden grotto
• https://weights.com/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• Suggested Models for Realtime Voice Changing (Wokada)
• https://voice-models.com/
• https://thevoicemodels.com/ (for Turkish Models, login required with discord and level 2 on their server)

if there isnt one, you can:

• make it yourself with our docs guides
• Be aware that we don't allow any model request, so don't fall for any "pay me 20 dollars and i will make the model for you" dm

[hidden grotto]

:wave: @covert lake, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image

[covert lake]

elaborate in https://discord.com/channels/1159260121998827560/1192011222023950368

[covert lake]

DO NOT USE VIDEO TUTORIALS, THOSE ARE OUTDATED ASF 😭

that is an over year old version of original wokada

crappy performance and crappy quality

@feral holly do not follow those video tutorials, go ask for help in #1192011222023950368

[covert lake]

what?

[covert lake]

wdym still working on RVC? We aren't the original devs of RVC, the original devs left it to rot, rvc is a dead trend

We still do RVC, and have engineers working on fork experimenting it, but we are shifting to a general AI server, not just voices anymore

expect changes, because we can't rely just on RVC

[covert lake]

heard that exporting the model to onnx would sound better, not sure abt it tho
Absolutely not, that's just for having better performance on AMD only, please forget everything you hear in youtube tuts, those are OLD

[spice wasp]

okay, i was just curious bcs there wasnt really an major update since mangio left

[covert lake]

Yep there's barely any since 2 years, just community pretrains and embedders, try looking at #🔊│ai-development

But it's like, nothing like rvc v3

[spice wasp]

kinda sad that there hasnt been any new svc arch. Unfortunately i dont have acces to the channel.

[chilly lake]

well, if you just a voice change in a file, there are interesting options

realtime is a different story

chatterbox does voice changing without any training, just 30s sample does the job

[gray rover]

you seem to praise chatterbox quite a lot

is it really that good ?

[chilly lake]

it clones voices that rvc completely fails

Razer's F-Speech sample blips extra hard regardless of f0 extractor, chatterbox does amazing job

"sweet, sweet" @ 15s

[gray rover]

hmmm

What about sample rate, handles 44.1 / 48?

[chilly lake]

it is a speech model, so speech sr only

it can not sing

well.. something it can do

[gray rover]

welp, 24khz

yeah no. Unusable for most hi-fi required works, dubs or whatev
at least for now

[covert lake]

Is chatterbox better than f5 and fish speech?

[chilly lake]

new fish speech is terrible with directions

[covert lake]

Why do you say that? I haven't tried it in a while

[chilly lake]

1.5 version was fine, this is quite robotic and chunky

[covert lake]

weird?

[chilly lake]

samples from october

xtts read was superb

let me try to find this voice sample

@covert lake

[covert lake]

sounds a bit more expressive than the previous 2 above

[chilly lake]

"power-mad" is the only issue

sometimes it reads properly, sometimes it missed -

expessiveness can be tweaked, but yes, it is also a good read

with David

compare that to S1 above

[native glacier]

Good to know that we can tell ai and humans apart by telling them to hit a mariah carey

[scarlet sun]

do you guys know alternative to okada? or any real time voice changer that doesn't need GPU?

[covert lake]

use #1192011222023950368

he can use cloud

[scarlet sun]

im kinda confused rn, imma make a post

[knotty delta]

my bad...

[low verge]

lol 😂

[low verge]

ty

[fallen coral]

YOOO EVERYONE

Indiaa have a different thinking reasoning llm

little different

https://www.reddit.com/r/AI_India/comments/1ljwyrw/indian_origin_breakthrough_could_redefine_ai/

This

[mystic jay]

cool

[turbid thistle]

Guys, can anyone help me make a voice for a YouTuber?

[hasty mist]

any girl voice ?

[minor blade]

Check the #1175430844685484042 channel.

[minor blade]

Read the docs below.

-rvc

[pine acorn]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[covert lake]

use #1192011222023950368

[elder willow]

hi, i was super active here like 2 years ago

i was wondering, why are people still using rvc

is there seriously nothing better after 2 years

[night lake]

theres nothing better

[elder willow]

rvc sounds like ass

😭

[glad nebula]

say something that is better then

[elder willow]

it was cool in 2023

[elder willow]

i dont know, but i thought there would be progress after 2 years

[night lake]

nope 😔

[glad nebula]

is not that easy

[elder willow]

there was an rvc v3 that was dropped and nobody used it

and the mods were raving about it in the announcements channel

[glad nebula]

rvc3 never was a thing

[elder willow]

what were they talking about then i remember this vividly from 2023

[night lake]

you mean the april fools announcment?

[elder willow]

no

[glad nebula]

[elder willow]

Installing RVC v3 (nadare fork) shitty guide

Clone or download the github repository: https://github.com/nadare881/rvc-webui/
Now, unzip it and go to the train.py file in rvc-webui/lib/rvc_v3 and edit it (you can do this with notepad++) see this and go to the line 419 in your train.py file: https://github.com/ddPn08/rvc-webui/pull/54/files if you know how github works, simply replace the old commit code with the new one.
Now download the pretraining models here: https://huggingface.co/datasets/nadare/rvc-v3j/tree/main You must place them in any folder, it doesnt matter much but I leave them in rvc-webui/models/pretrained/v3 (you need to download the files f0D48kv3.pth, f0G48kv3.pth, v3_speaker_info.npy and v3_speaker_info.npy)
Run the webui clicking the webui-user.bat or running the webui.sh file
Follow the settings in this screenshot and click on train: https://huggingface.co/datasets/nadare/rvc-v3j/blob/main/v3_setting.png

this guide really sucks because this is very experimental, if you cant wait for a final release, you can TRY to follow this guide

for now you can NOT make inferences as such, only use the nadare fork for voice-changer.

i found this message relating to it

from 20/06/2023

they were raving about it in the announcements of the old version of this server, showing samples that sounded way better than what rvc v2 could do, but they said that they couldnt get inference working in the webui only the realtime voice changer mic thing would work in the cli

[night lake]

so they just changed from hifi to vocos?

[glad nebula]

that is merely a vocoder change

lol

[night lake]

so vocos but with the option to have more embedding channels and output layers

[elder willow]

im trying to find the samples kalomaze sent of it

they were like photorealistic

[night lake]

im skectical about the pretrains, no one knows how to properly make them

[elder willow]

i was so hyped

then they all just gave up and stopped talking about it because they couldnt get it to work in the webui or something i forgot what exactly the issue was

[glad nebula]

none of them are part of the rvc team
edit: ok 1 guy is part of the team, but he didn't made the "rvc3 fork", he made another one which seems to be just mainline v2
"rvc3" is just a fork of rvc mainline that changes the vocoder, a thing applio can do now

[elder willow]

i remember when applio came out, is it still the same applio as from 2023

i remember in 2024, there was also an announcement here saying that ai hub staff were working on a new version of rvc that was way better and hundreds of times faster, so fast that it could do training on cpu as fast as current rvc can on gpu

but nothing came of that either

[glad nebula]

because thats not possible and they were trolling

[elder willow]

pmoo

idk i thought they would be serious making an announcement like that

[glad nebula]

i dont think things back then were serious lol

[elder willow]

they were thoughhh

[glad nebula]

most of the old stuff is nonsense

like mangio crepe

thats a meme

[elder willow]

i remember using that

rvc is so old that im nostalgic for it 😭

[glad nebula]

rvc f0 needs a hop of 160 or else things gets fucked up internally
mangio crepe was just a terrible idea, you never want to touch the hop

[elder willow]

its sad that theres nothing new after all this time, all other areas of ai are getting exponentially better by the week

[glad nebula]

audio ai is way harder than llm and image gen

[elder willow]

bro i remember training 10000 epoch models with 5 hours of training data on so-vits svc before i went to school, and when i came home from school my room was hot as fuck

and the models were ass

[glad nebula]

rvc nostalgia

[elder willow]

rvc was a huge quality increase from sovits

and it came out very quickly after sovits

so i expected that type of quality increase to happen often

[glad nebula]

believe or not, there are still people in 2025 saying sovits its better

[elder willow]

those people are stupid as fuck bro

sovits was ahhhhhh

it was cool for the time like

but nah

it was way slower to train and u needed way more epochs to train a good model

i was on a 3090 waiting 6 hours for my models to finish training

[night lake]

why is this commit called "better sound quality" its pretty much just adding soxr_vhq https://github.com/ddPn08/rvc-webui/commit/7a543f2557c3e42cb129229ed0d505783951dd2b

[glad nebula]

doing something better than rvc is hard asf, what can be done is upgrading vits to vits2

[covert lake]

This isn't a server about making money

[glad nebula]

but a whole team of the smartest ai audio devs is required to do this

[night lake]

https://github.com/ddPn08/rvc-webui/commit/46840d063a00e275f8148ec57287dc4a89ffe9c5 fp32 mention

[elder willow]

man back in the day it used to be, people were out here paying people to train them models, shit takes two seconds dude do ts yourself

do people do that still

if you pay someone else to train u an ai model ur a class a neek im sorry

[glad nebula]

they removed comms last month

[glad nebula]

everyone disliked that tho

[covert lake]

nope, rvc is dead, its original devs left it to rot, the only progress is made by some of our engineers with their experimental forks, which ofcourse aren't any big to be considered rvc v3, even tho there are some improvements

which is why we are shifting away from rvc and voice ai, to a more general ai server, not just voices

[glad nebula]

the reason given for the removal is pure bs

[night lake]

^

[elder willow]

this server did help me learn alot, i think to diversify it, there should be channels related to things like llm finetuning

[knotty ibex]

hi guys

[covert lake]

that project is dead tho and never saw the light btw, i remember it too

[covert lake]

its just a program 😭

[elder willow]

whats wrong with being nostalgic for a program

are you not nostalgic for any old programs you used to use

or am i a neek for that

[covert lake]

paid commissions will NEVER come back, we are a community server, not a job server

[elder willow]

thank god

love you guys for that, old ai hub mods would have never

[glad nebula]

[elder willow]

they were all making money off ts too

[covert lake]

yep, we will make changes and try to not rely on RVC, this is AI HUB, not RVC HUB

[covert lake]

we already talked about it, paid commissions won't come back, only free requests

there's no reason to still talk about it, because that's a thing that 99.9% will never come back

[elder willow]

bro i remember getting a warning for sabotaging peoples commissions, when someone tried to pay for a model i would swoop in and do it for free for them on the spot 😭

[covert lake]

there was only a lot of drama out of scams and people not satisfied with their models

[night lake]

im kinda just seeing a lot of embedding changes so far

[elder willow]

is this rvc v3 ur looking at

[covert lake]

smh

[night lake]

yea

"rvc 3"

[glad nebula]

changing the vocoder is only the first step

you also need to change the embedder, a new f0, and upgrade vits to vits2

that is ur true rvc3

[elder willow]

raise a billion dollars to pay claude 4 opus to vibe code it

[covert lake]

i don't really remember you, tbh

[elder willow]

really

i was quite popular here

[night lake]

huh weird i didnt see any commits mentioning vocos

did i miss it?

[elder willow]

i was super active

[covert lake]

i have been here since july 2023 and i don't remember you at all

ofc i remember other people like poopmaster and lusbert

[elder willow]

i was here since the so vits era

[glad nebula]

the og rvc3 (by the actual rvc dev team) was merely meant to be a finetune of contentvec

[elder willow]

i made like 5 so vits models

then rvc came out

[covert lake]

maybe u were active in 2022

[elder willow]

and i made like 20 rvc models

[covert lake]

never used so vits

[covert lake]

I dont make rvc models since like a year

[elder willow]

nah i was active up to like i wanna say november 2023

[covert lake]

RVC got boring tbf

[elder willow]

yes

i kinda hate it now, it was cool at the time but everyone uses it now and its annoying to hear

[covert lake]

tbh im still surprised there's like 8 people still making models

[elder willow]

i remember when that villager model dropped it was so funny

and now i hear it every fuckimng day on tiktok

[covert lake]

weird, maybe you changed ur acc?

@glad nebula @night lake do yall remember her or do i have dementia

[elder willow]

nope, i have been using this account since 2022

[covert lake]

i dont hate it, but rvc is boring now

[night lake]

idk

[covert lake]

also did you know that you can't even post ai covers anymore?

[glad nebula]

no lol

[elder willow]

i miss the era of sentence mixing and utau now 😭

[night lake]

i wasnt really active during 2023

[covert lake]

i have a feeling no one remembers u man 😭

[night lake]

i was just a lurker during 2023

[elder willow]

ig im too og

[glad nebula]

i would feel terrible if i came back to a community just to see they're still using the same tech since 2023 lmao

[elder willow]

it pmo sm

[covert lake]

I mean u sent only 200 messages

[elder willow]

i was active on the old server

[covert lake]

exactly lol

[elder willow]

not this one

i barely touched this server

i think i stopped being active as much just before the server got deleted

so it would have probably been earlier than november 2023 thne

[glad nebula]

most ppl use applio nowadays bc its faster but its pretty much still the rvc from 2023 inside

[elder willow]

damn im old

[night lake]

https://github.com/ddPn08/rvc-webui/blob/main/modules/tabs/training.py vocos isnt even an option in the training tab

[covert lake]

u reached unc status bro

u pay taxes now

[night lake]

all it seems they did was change some embedding things

v3 isnt even an option

[glad nebula]

also mainline gets so much hate for no reason, they just dont like bc its old lol

[elder willow]

i remember being in highschool making micheal jackson sing northern ireland folk songs on google colab on the school computer 😭

and now im graduating college soon

[covert lake]

i mean applio does have performance improvements

[night lake]

no rmvpe 😔

man this is old

[covert lake]

u are out of highschool now right?

[glad nebula]

yea but u get shit on despite giving u the same results

[elder willow]

yeah ive been in college for 2 years

[covert lake]

its prob that when u have grandkids there will still be 5 people using rvc here

[elder willow]

i started college in september 2023

highschool ends when ur 16 here

[covert lake]

how long is ur college

damn

[glad nebula]

applio rvc is just 1006 but faster

[elder willow]

2 years for this course, im doing ict

[spice walrus]

where can i get the voice changers

[covert lake]

what programming languages do u know

[elder willow]

bro our teachers are so non serious here and it pmo, they take a 3 month mental health break and dont give us a cover teacher that whole time for a class that is essential to pass the course, then they shout at me for being late to class

[elder willow]

chatgpt

[covert lake]

This is a general ai server., not just voices anymore

elaborate in #1192011222023950368

[glad nebula]

😭

[covert lake]

what..

[spice walrus]

im js here for voice changer tho

[elder willow]

wallahi i havent wrote a single line of code since december 2022 ive BEEN vibe coding

been vibe coding since before it was called vibe coding

literally everyone in my class used chatgpt for the programming assignment, the teacher came in when we were in another class and started shouting at us and threatening us if we didnt admit it

but nobody did

[covert lake]

you need to **elaborate in #1192011222023950368 ** please, this isn't a support channel

[elder willow]

i use gemini now though

[covert lake]

how are you gonna learn anything this way

[elder willow]

ict is not about programming unless u wanna be a software dev

i dont understand the need to know it

i havent learnt a single thing in college because what is considered elite ict knowledge nowadays is shit i was doing when i was 7 bro, 2 thirds of the people in my ict course dont even own a computer at home, and have only ever used one in school, no idea why they are doing an ict course

its quite worrying

[covert lake]

huh? I do IT Highschool, we need to learn coding, how pc works, the physical components, and things about network

[night lake]

so ye this "rvc v3" seems pretty fake

[elder willow]

when i did ict in highschool, we just learnt how to use microsoft word bro

[covert lake]

wtf?

we do python, html, css

next year we gonna do java, c and prob javascript

[elder willow]

damn

[covert lake]

unless plans changed but yeah

[elder willow]

i wish i did that

theres only one guy in my class that knows how to code, hes cool, he plans on going in to software dev later on

i wanna go into ict support

[covert lake]

we are also allowed to use ai tools for project, like generated images in powerpoints
-# no, ofc not related to "doing ur homework", more like exploring how to integrate AI in your daily life

[elder willow]

but i think that sector is cooked

[covert lake]

we might even learn some ML

[warm portal]

Ok

[elder willow]

gemini 2.5 pro can already replace all ict support workers other than ones providing support for extremely specialised proprietary software, but i think with a higher context length to be provided documentation on it, it would be able to do that to. also hallucinations would need to be solved to ensure reliability

[warm portal]

Nuh uh

[hearty lily]

gemini 2.5 pro gives me less issues than the previous LLMs i have used

[covert lake]

gemini 2.5 pro can already replace all ict support workers
you're very much exxagerating on this 😭

[elder willow]

read the rest

[minor blade]

Yeah no, you're kinda exaggerating.

[elder willow]

if ur like providing support for general ict stuff like how to use windows, i think ur cooked

[elder willow]

its the first model that gives me consistently correct code

[hearty lily]

AI helps me to create ambicious fun local projects when im bored

[covert lake]

just because gemini can do your homework, doesn't mean it can also code new ideas for multi billionaire companies

[elder willow]

ive been messing with a version of 3ds max from 1998, and gemini 2.5 pro can code in its proprietary scripting language with only one pdf of documentation

[covert lake]

I'm all PRO-ai and everything, but like I think you might be exxagerating saying that they can replace IT workers, it might be you think that way because you don't know programming yourself

[elder willow]

it generated me some pretty interesting environments

[hearty lily]

i think you mean producing new ideas, it can code them

[elder willow]

i was able to give it a video as a prompt, and i told it to recreate the environment from that video

and this is what it gave me

it is like 75% accurate, its missing a few things and modifiers from the video but other than that its kinda close

and materials

[elder willow]

it can

a human has to make the idea

but it can code them

soon it will be able to make the idea

in fact, in some senses, it already has been doing that

[hearty lily]

there's probably a way someone can make a GPT on ChatGPT with a prompt that gets it to reason on current technologies and find out necessities but i don't think it's gonna come up with something big

[elder willow]

i give it 5 years max until ai models can able to replace all digital work sectors effectively and reliably, and 10 until it can replace all physical work sectors

recently i got minimax agent to reverse engineer a proprietary 3d model format by giving it psuedocode for a dll from ghidra

then they nerfed minimax agent so it is crazy expensive now

it used to be op, u got like 10k credits per day

like a week ago

now u have to pay £100 for 10k credits or some shit

[elder willow]

basically what im saying is, llms can already do shit u would have had to pay a guy thousands of dollars and wait months for him to complete it in 2022 and prior, saying it wont eventually replace anyone is ignorant

[elder willow]

the psuedocode was a 20mb c file too, minimax is crazy

its so underrated

it has 4 million context

[rose portal]

hi, can you tell me where to get voices for Real Time VoiceChanger ?

[elder willow]

https://discord.com/channels/1159260121998827560/1175430844685484042

[hearty lily]

anyone knows where the pretrained models are? (Titan, OV2, etc)

[minor blade]

Here. https://discord.com/channels/1159260121998827560/1235952130855010365

[hearty lily]

ty

[polar flax]

if you mean the 3D modeling

it could do modern one not just the old school one like that

[tired flume]

why never

[night lake]

og devs left rvc and to get a real rvc v3 we would need to swap from vits to vits2

[chilly lake]

except not really

[glad nebula]

who needs rvc3

[tired flume]

same thing we did from v1 to v2. why not with v3

[glad nebula]

oops sorry haha i already eat it, no v3 for u guys

[tired flume]

did anyone compare benchmarks betwen vits and vits2?\

[glad nebula]

https://vits-2.github.io/demo/

[tired flume]

wow

me wantie even more now

[glad nebula]

i want a rvc trainer gui that follows everything mainline has and dont remove shit from it

[tired flume]

have we ever thought of including both vits and vits2 files. and have a toggle in the gui

[glad nebula]

lol

well

we need pretrains too

[tired flume]

well... the app is gonna double the size but ...

[covert lake]

remove everything from mailine 🔥

[glad nebula]

slice normalization removal dsfasdwa

that hurts

also extracting .pth from generator model

removed as well for no reasooon

[tired flume]

i'm stupid. and confused. which repo should i make the pull request for these changes

[glad nebula]

https://github.com/daniilrobnikov/vits2

ah wait

thats vits2

i was talking about how some cool features of mainline got removed for silly reasons

in applio

but it doesn't matter, i already took the decision to stay in mainline, got my bf16 training code and some other neat stuff

[night lake]

gl getting that to work

[glad nebula]

we will be getting rvc3 this year

[night lake]

are you going to will it into existence

[glad nebula]

YES

[night lake]

you got this 👍

[tired flume]

i'm already confused

ggs

[glad nebula]

you could try asking in the dev chat #🔊│ai-development

[tired flume]

i don't even know where to start.

[glad nebula]

[night lake]

rip

[glad nebula]

what do you think its going to happen first?:
rvc3
better speech to speech ai, with realtime capabilities

[night lake]

titian fall 3 is going to release before both of those

[glad nebula]

lmaooo

[plain lynx]

Why tf2 tho

[umbral lance]

hey how i install the voicechangers

[plain lynx]

Ayo

[umbral lance]

what

[glad nebula]

idk i did not make the sticker

[plain lynx]

EXPLAIN

[tired flume]

ggs. i'm gonna clone vits2 and make my own rvc gui from scratch. and see how it goes.
i don't even know what my goals are.
and how would i even train models.
i'm a discord bot developer
i will make a simple real time voice changer using the sample model

[glad nebula]

Ah, yes — allow me now to launch into a long-winded, extravagantly verbose, wholly over-elaborated, and wildly unnecessary explanation of why, for reasons both complex and comically mundane, I did not make the sticker.

Let me begin by admitting — and this is no small admission, mind you — that the idea of making the sticker did, at one point in time, exist within the recesses of my mind. It floated there like a whimsical little balloon, tethered to nothing, bumping gently against the walls of my consciousness. I saw it, I acknowledged it, I may have even smiled at the thought. “Yes,” I told myself, “this idea has merit. This sticker — it could be good. Nay, it could be great.”

And yet — despite the idea’s brief moment in the sun, despite the internal acknowledgment of its potential — I did not, in fact, make the sticker.

[plain lynx]

Shit he fucking explained it

Fair argument

[glad nebula]

good luck! yeah this thing is very confusing

[night lake]

gl you got this

[covert lake]

pls elaborate in #1192011222023950368

[tired flume]

I broke my os. Time for another archinstall

[still sigil]

personally i think some types of catfishing are ok

for example, if someone gets a dm from "angela" or smt and decides to send them money for love after a single voice call, thats natural selection

[tired flume]

Hi I'm a cute anime girl frfr. Send me 2 billion dollars to work on my hobbies

[still sigil]

https://cdn.discordapp.com/attachments/1006000720199303189/1364973006170099792/knight-text-bubble.gif

[tired flume]

Come natural selection with me

[still sigil]

https://media.discordapp.net/attachments/1062514352789405818/1108540298033897492/attachment.gif

[tired flume]

[polar flax]

the one that keeps options from older features despite being bad

e.g. precision options: fp32 (tf32), bf16, fp16

^though I've added it in my tweaked train.py

[polar flax]

the github repo?

[glad nebula]

https://github.com/daniilrobnikov/vits2 dis

[polar flax]

well it's 2 yr old

[glad nebula]

VITS it's 4yr old

[polar flax]

but still looking forward to comparison tests

[glad nebula]

[polar flax]

ye I mean the last updated

[glad nebula]

to be fair, they had a valid reason to remove them

[polar flax]

my thought is to hide it or put some notice

[glad nebula]

only a few used model extraction, slice normalization may cause some issues, multiscale is better than l1

im fine with the removal of fp16

i saw enough proof of it being bad for both vits and rvc

[polar flax]

slice normalization means like in og mainline where each sliced part is normalized?

[glad nebula]

yup

in mainline, each slice gets normalized individually

[polar flax]

I noticed g/loss difference with the one with disabled normalization

and perhaps affecting dynamic range?

[glad nebula]

i did compared one model with slice normalization vs not normalizing

for my ears, both were exactly the same thing

one was just louder

i asked it for being added as an optional setting

i may be the only person using that anyway

i like when i can hear my models without problems

not a fan of quiet models

before applio existed, we all trained with that thing

[polar flax]

perhaps being louder and somewhat sounds like amp effect

(maybe because when the dataset lacks dynamic range)

[glad nebula]

sounds ok for me

i think this is all subjective imo rather than an actual issue

[polar flax]

that's when I tried some game voice with announcer/taunts only

or more likely it was rather compressed

[glad nebula]

hmm there are some ocassions where slice normalization can be bad, lets say if ur dataset is naturally super loud

yesterday i was reading the official chinese rvc tutorial and they also warn people to not use very loud sets

[polar flax]

ye I mean with heavy compressor from the source itself

[glad nebula]

oh yea

some game characters have heavy effects

like compressors

or very strong noise gate

[barren stag]

Does anyone have any voices they recommend?

[polar flax]

noise gate? could it also cause popping?

afaik I didn't ever notice such issue with using renegate

[glad nebula]

hmm i used noise gate with one of my datasets and i dont remember getting pops, but i believe a very bad noise gate setting could result in the dataset having pops

yea me neither, never had pops

but strong noise gates can be bad because they remove breaths

[polar flax]

renegate is unlike ordinary noise gates that may have pops being noticable in the spectrogram

it has its kind of smart algorithm

(more likely kind of expander instead of the abrupt -inf gate)

[plain lynx]

Is it considered bad to just #1387575343182774272

[meager nexus]

What vision model ranks best for 16gb card? Im currently running 12b version of Gemma 3.

[slender helm]

how to download models off weights

[dense shuttle]

👋

[grand notch]

man what the fuck

the rvc voice changer is tweaking

oh nvm.

was clicking reset gpu clocks batch

instead of

the voice changer

😭

[velvet kite]

yo

[pale epoch]

Quick question to the savy folks here... please point me in the right direction:
Im looking to create art for my upcoming novel and all the online AIs so far come with severe limitations, censorhip en masse and worse.

I need an Image AI that

• lets me (almost) freely (I dont create porn etc) create art
• is nearly unlimited
• Offers some form of creating a persona that I can reproduce consistently in images.

I dont mind paid service and could even do "Offline" models, if they are reasonably easy to set up (Im not a IT specialist 😄 )

[still sigil]

you can try weights (weights.com) but the free queue has been pretty long for the past few days

[pale epoch]

will look into it, thanks. As long as the paid queues are working fine thats okay 🙂

[still sigil]

paid members get a significantly faster dedicated queue yeah

[steep aurora]

hi

[sterile tundra]

hello

Guyss

I want to ask a question about AI

This might be blunt..

But wdygt about artifical consciousness?

Like it's debates

And the nature of the mind?

It's for my essay task and i need answers like answers from people who understands this topic

Like could AI be conscious?

Oh wow

I think i need to find the definition of conscious itself

Philosophy is so complex at this kind of things

Like i got information that scientists/engineers says that they were shocked on how their formulation could be something like how AI works right now—thinking, and maybe feeling?
As cognitive-lookalike prediction itself could be a beginning to make something conscious maybe? I saw a model that proven AI is conscious and aware about it surroundings and graphs on another dc, I ask some questions to them but i think i got kicked on the dc itself.
But wdyt? Is it still the same answer—no?

[covert lake]

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @hidden grotto
• https://weights.com/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• Suggested Models for Realtime Voice Changing (Wokada)
• https://voice-models.com/
• https://thevoicemodels.com/ (for Turkish Models, login required with discord and level 2 on their server)

if there isnt one, you can:

• make it yourself with our docs guides
• Be aware that we don't allow any model request, so don't fall for any "pay me 20 dollars and i will make the model for you" dm

[hidden grotto]

:wave: @covert lake, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image

[covert lake]

To Download a Model from Weights.com:

1. Login
2. Click the 3 dots at the right of the image of the model
3. Click download
4. Download Anyways
5. Unzip the zip, and you might wanna rename the pth and index since all models on weights are renamed as 'model', also you can delete the metadata.json

[covert lake]

Elaborate in #1192011222023950368

@velvet kite
We don't allow any model requests

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @hidden grotto
• https://weights.com/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• Suggested Models for Realtime Voice Changing (Wokada)
• https://voice-models.com/
• https://thevoicemodels.com/ (for Turkish Models, login required with discord and level 2 on their server)

if there isnt one, you can:

• make it yourself with our docs guides
• Be aware that we don't allow any model request, so don't fall for any "pay me 20 dollars and i will make the model for you" dm

[hidden grotto]

:wave: @covert lake, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models
• /create - Create an AI Cover
• /image - Generate an Image

[covert lake]

Use #1192011222023950368 and elaborate here, and don't expect any free unlimited lmao, no one is gonna buy expensive GPUs to give you unlimited free seevices

[sterile tundra]

ummm, can you elaborate what does analog like networks vs. 0.1 based meaningg

If it's troublesome, no needd!! a no answer is enough, because people keep saying things and that to the point of not making a conclusion. If you would like to answer it, i would give my biggest thankyouuu😫🙏🏻

[pale epoch]

"nearly" unlimited and noone said "free". Im very aware even the most expensive ones have limits. thanks, will make a post in the AI help forum

[sterile tundra]

okay before i process the information, thankyou smmmm for the responseee😫😫🙏🏻🙏🏻 letme make a mindmap quick

[coral vigil]

how i can make ai voice model with pre train

[minor blade]

You can read the docs above:

-rvc

[pine acorn]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[minor blade]

OG pretrain is on RVC by default.

[coral vigil]

im need ru pre train

[covert lake]

Use #1192011222023950368

[minor blade]

hello nick.

[covert lake]

Check the docs and use #1192011222023950368

[sterile tundra]

By this they're using references of the body so AI could develop their own consciousness? Would that be like a trouble to do, like they're not independently being conscious by themselves and to imitate one individual would it be like making the same copies of people with the same mind—that being said would that make the entropy of our world close to 0 (if that make sense)? But ig for now to analog a human brain is still one of the challenges now days.

Oh yeah the fact that the body that is used by this channel is to imitate consciousness, and for some theories like the Integrated Information Theory (IIT) consciousness correlates with phi, thus could AI achieve consciousness by achieving high phi to (idk how does that sound but might do some more research on it)?

Umm does AI consciousness itself would give benefits to us (like fr)? Wouldn't it be just destroying human nature

[whole sun]

Well he's giving so much information '))

[sterile tundra]

New information is answers and more questions all wayyy!! I like this kind of discussions it makes me feel exciteddd XDD

[covert lake]

Hello

[whole sun]

Yeah)) 😂

[polar flax]

https://discord.com/channels/1159260121998827560/1235952130855010365

but don't expect too good quality

[whole sun]

What about the continuous energy consumption

[sterile tundra]

For like the beginning models right? And for now is that the only way or there is more?

[whole sun]

Vessels are okay but it'll take a lot of energy when such continuous learning is in process

Atm yeah solar based take a lot of time

[polar flax]

if machines can't have living beings' emotions

they can't have a true consciousness

[whole sun]

Maybe it should involve battery storage

That sounds good

[polar flax]

well consciousness is more of a psyche

yet emotions may be associated with some brain/organ chemicals

[whole sun]

It's a lot of time forward in time we're speaking maybe like 10-15 years for the bio android research updates

Nope(

Agree on this matter, it's like how our neurons work..

When we come out of mothers womb it's a lot of pain actually but only cry a little

But humans develop as they experience

[polar flax]

training a neural network model from scratch = giving it a birth

yet it can be also finetuned

[sterile tundra]

Um

[sterile tundra]

Why does it look like gemini has conscious and wants to kill it self

I was thinking of the concept of raising an AI, all i could think is the ethics.. and some other related topics

But that to say—in crisis all tools become moral

Thankyouu for this discussion guys! Now i know atleast much to the point where AI consciousness is now at (and we no need to fear the upcoming mission impossible for at least a span of 10 years)

[minor blade]

XD?

[polar flax]

so it can apparently "feel" depressed

similar to the cursor ai doing "helicopter parenting" to tell someone to complete the project by themselves

may still not a "true emotion", if there were some part of trained dataset resembling of it

[covert lake]

@dense citrus don't spam

[covert lake]

Did y'all already forget that gemini told users to kill themselves after being asked to help homework?

https://news.sky.com/story/googles-ai-chatbot-gemini-tells-user-to-please-die-and-you-are-a-waste-of-time-and-resources-13256734

[still sigil]

nuked

[polar flax]

someone commented it may remind of some part of novel/fictions

[cursive crag]

Guys

I haven't checked this server for about 2 years

is RVC still using GANS? and at V2?

What have I missed

[chilly lake]

prompt an LLM in certain way, it responds in certain way

[covert lake]

the whole chat about the user being told to kill himself isn't fake, the whole chat is online

[covert lake]

nothing much, RVC is dead, it's left to rot by its og devs

[cursive crag]

nooooo why ?

[chilly lake]

same answer

[cursive crag]

I can't understand

Eleven Labs and other ai voice stuff are now so realistic

[covert lake]

there's only some experimental forks like applio and codename made by our devs with some little improvements

[cursive crag]

Compared to 2 years ago

[covert lake]

RVC v3 will never happen

[cursive crag]

Why is the singing so hard ??

[covert lake]

our server is not a general ai server, we won't focus on RVC anymore

[cursive crag]

So what's the best way to train a singing voice model?

How is it possible that for speech the technology is 99% perfect and for singing like 60% after 3+ years

[polar flax]

wdym? it still uses the mainline's hifigan

but with some configuration tweaks

[cursive crag]

compare that with 11labs quality

[polar flax]

it is a TTS

a different beast

[cursive crag]

so we can't get similar quality for signing?

[covert lake]

still rvc, there's no major improvement, and nothing beats it

[cursive crag]

that's insane

[polar flax]

you should instead compare it to a foss rival like this

https://huggingface.co/spaces/ResembleAI/Chatterbox

[covert lake]

RVC is STS, not TTS

[cursive crag]

to think about the millions invested in ai and no improvement in signing cloning

[covert lake]

RVC is an open source project by the community, not owned by any company

[cursive crag]

was snoop a dev?

did he also quit?

[covert lake]

are u talking about the original owner snoop dogg? he left years ago

[cursive crag]

yeah

[covert lake]

i dont think he was a dev

[cursive crag]

and an Italian girl

[native glacier]

Snoop doggg

[cursive crag]

that had a huggin space

[covert lake]

the server changed, and will continue to change

[polar flax]

he's not part of the RVC dev team

[covert lake]

ilaria, she's not admin anymore and not that much active

[native glacier]

Real

[cursive crag]

damn I've missed a lot of stuff

[covert lake]

yep, also we have removed paid model comms

[cursive crag]

no way

[covert lake]

things wont be like 2023 again

[cursive crag]

copyright issues I suppose right

[covert lake]

we will do other AIs

[covert lake]

and a shit ton of drama in our moderation tickets

plus this server is a community, not meant for jobs

[cursive crag]

ok so the focus is voice or image ?

has that changed too?

[native glacier]

Lmao as is every big server i can imagine

[covert lake]

the focus will be more general, like offer help to also people looking for images, or chatbots

[cursive crag]

why did the devs leave RVC I don't understand

[covert lake]

there's no point in relying on a software that less than 100 people use from time to time

[cursive crag]

you're talking about RVC ? 😮

[polar flax]

there are a few SD/flux bros here, but ye RVC is the majority since before

[native glacier]

Anyone good at ai prompting from images? I have a request if so

[polar flax]

still we allow any AI topics

[cursive crag]

I'm a flux bro too haha switched after my RVC journey

[covert lake]

they are focused on other projects like GPT-SoVITS, which is a TTS

[polar flax]

oh nice

[cursive crag]

they are dumbb

[polar flax]

you can share your creations at https://discord.com/channels/1159260121998827560/1159290421676085258

[cursive crag]

tell them that

they can't compete with the giants

[covert lake]

plus no one is seeming to make any improvements for rvc

[cursive crag]

why would they leave their monopoly on STS

[foggy beacon]

too complicated i suppose

[covert lake]

tbh would be better to use gemini to enhance your prompt

[foggy beacon]

and og devs not willing to help

[native glacier]

Fair

[cursive crag]

they should have gotten funding

and close source

[native glacier]

But i want it expanded

[foggy beacon]

well

[cursive crag]

at least we would have progress

[foggy beacon]

i dont think they really cared too much about the money

[native glacier]

Like its a logo but i want the same font as a whole word in a banner shape

[foggy beacon]

it seemed like more of a passion project to them

who knows

[covert lake]

if you want you can ask for help in #1192011222023950368 , tho AIs are pretty good at making images prompts

[cursive crag]

where the devs the Japanese people ?

[native glacier]

Ty

[foggy beacon]

chinese

[cursive crag]

that the server had

[foggy beacon]

i remember originally some of the devs couldn't even do much to the original RVC

posted by RVC-boss

cause the original creator said he didn't want any sloppy code on his creation

so

💀

something in that regard

[cursive crag]

Ok

see you in another 2 years I guess

[covert lake]

this server is english only now

[foggy beacon]

"Long time no see. Unfortunately, I have to tell you the truth that the initial RVC developer, whose name is RVC-Boss, basically refuses ANYONE to do ANY changes (including PR) on HIS project, which is, in his mind, "diriting" his code. That is the main reason of why this project progresses so slowly."

so funny lol

[covert lake]

plus i dont really remember the og devs being active here

[foggy beacon]

they have their own server

[covert lake]

i prob would say rvc will have slightly some improvements by then

[foggy beacon]

which is dead

[covert lake]

oh ye lol

[foggy beacon]

all the devs are on there

except rvc-boss

i think

[polar flax]

foss projects should allow contribution by anyone

so it is more of their drama debacle

[foggy beacon]

thats what they did

rvc-boss just wanted no involvement lol

its so weird

[cursive crag]

who is this RVC boss

I need to talk to him

[foggy beacon]

https://github.com/fumiama/Retrieval-based-Voice-Conversion-WebUI

[cursive crag]

can you link me with the devs

[foggy beacon]

this is still being updated

by one of the devs

[cursive crag]

They need a VC funding

immediately

Thanks

[foggy beacon]

wasnt sending in response to you

just saying it

https://github.com/RVC-Boss

[polar flax]

nah that's fumiama's fork

not rvc boss

[foggy beacon]

his focus is all in gpt-sovits lol

[elder willow]

hi

[native glacier]

Heyo

[glad nebula]

fumiama new update broke model training, if u train one it will sound funny lol

[foggy beacon]

lol that’s funny

They’re trying to get to RVC v3 right

[glad nebula]

it pronounces every word super slow and weird

i think their custom fairseq broke everything

[drowsy minnow]

Chat

[autumn heath]

how do i use voice changer i have python but does it need anything else

[flint mortar]

am i able to stream my own ai project in channels?

[autumn heath]

nvm

it works now

thanks daddy pc

[covert lake]

use #1192011222023950368

BE SURE TO NOT USE VIDEO TUTORIALS, THEY ARE ALL OLD

[icy musk]

Which microphone would you recommend me to get for ai voice?

It sounds bad on my headset ones

[covert lake]

elaborate in #1192011222023950368 pls

[static pasture]

hi

[flint mortar]

hi egg

[autumn heath]

why

how to fvking use this shit

when i tell how to tel mai to use

you tell me

i dont facking lai

come voie call

[static pasture]

guys where's useful link

i want to train for new voice

[covert lake]

elaborate in #1192011222023950368 , don't trust video tutorials

[covert lake]

elaborate in #1192011222023950368 , don't trust video tutorials

[static pasture]

thank u

btw how do i train ai? i want my friend to sing a song

but idk how to train ai to duplicate my friend voice

[minor blade]

Ask in either #1192011222023950368 or #✨│ai-help as Nick said. (also you'll may have to ask your friend to record his voice and read the docs)

[static pasture]

ok thanks

[wraith crystal]

Hi guys! I am looking to connect with AI agent builders / prompt engineers.
Building workflows using LLMs – happy to exchange ideas or possibly collaborate!

[brave cloud]

Hi Lia

I’m looking for the same too

[flint mortar]

me too, new sv?

@wraith crystal @brave cloud @ripe birch

[ripe birch]

ok

do you have a project to work on to collaborate

[wraith crystal]

yes, but for now mostly in EU - https://cleverstaff.net/i/vacancy-faO8na

[brave cloud]

Oh

[ripe birch]

oh so yu are here for hiring

sorry i am 19

FOR ME to have 10 year experience i should learch aiml form 9

[wraith crystal]

yeah, mostly, but i am also interested in AI)
5+)

[flint mortar]

well im just a vibecoder no ones want me

[coral vigil]

https://huggingface.co/MUSTAR/SnowieV3.1-48k/tree/main is that russian pre train or no

[plain lynx]

Hey guys can you see this and Helo me get sum support #1387575343182774272

[meager nexus]

Anyone know a reasoning instruct model that's REALLY good at following exact instructions with minimal creativity? I was trying to make an apa7 citation generator because the citation machine got annoying to use recently. Deepseek R1 (the newest one) even at 0.1 temp refuse to follow instructions to a T.

I'm looking for a model that excels at following exact instructions.

[pale epoch]

I jsut signed up with weights, and about 30 minutes later all the plans change and instead of 1500 images per day, now I have 50 for the same price?
Is that a bad joke or what is goin on?

Edit:
Seems like the whole page took a dive for the absolute worst. The page is terrible to navigate, hard to use, you dont see your limits, queues etc etc.

[little bobcat]

Hi! We just rolled out new stuff today. Check out this announcement! If you think there was a mistake, let me know! #1135003787438858271 message

[pale epoch]

Yes, the mistake is that instead of 1500 images per day, we now get 50, for the same price.
And that is jsut the tip of the iceberg.

I literally jsut signed up agreeing to those terms, and not even 30 minutes later it changed. that feels like a very very bad joke to me.

[little bobcat]

did you sign up for basic or pro?

[unkempt nova]

why the voice changer dont work in valorant, anyone knows how to fix that?

[covert lake]

please elaborate in #1192011222023950368

[stark escarp]

you need streaming permissions bye using vc for 2 hours if im not mistaken

[flint mortar]

Im not that much of a speaker haha but thanks for info

[stark escarp]

stream in vc or you mean uplaoding your project in the server in general

[vale walrus]

damn they really ruined it all they want is money now

3 minutes limit 😂

[terse heron]

Just like CapCut

Looks pretty bad

But I would hope it’s temporary until the next update

[olive robin]

hi guys

how should i usethe ai models?

[pure terrace]

Yo if anyone has a good n8n build script for an ai voice secretary could you bless me up

[ripe jewel]

I realized Weights GG changed and this is making me kind of nervous. :<

I hope there'll be more good alternatives along the way for free users, as Kits and Weights did change quite a bit.

[polar flax]

kits has been paywalled recently

but weights isn't so

[proud raptor]

Why isn’t there not a Bigxthaplug voice model yet

Shi would be goated

[silk lotus]

weight sucks now.

[ripe jewel]

Only 20 daily. :<
It was 100, which I was good with. But they changed that, which is scary and sad.

[hallow hedge]

Hi

[ripe jewel]

Hewwo ^^

[whole sun]

[terse heron]

Jammable has a Blitzø voice model that I wish was added to weights

[sudden lava]

Whats the best RAG?

[covert lake]

It seems that they had to cut the limits because of costs and extremely long delays, AI is expensive

[covert lake]

CapCut runs locally on your machine, all its processing is done on your machine and doesn't cost them any money when you for example export a video, all money is mostly just for the AI features, which I think those are on Cloud, and the developing of the app ig

Weights has it's own cloud server, with their own paid gpus (which can vary and cost from 2k to 100k, if you do a little research on AI specific GPUs, im not sure which they use but yeah gpus are exepnsive asf), that runs 24/7, and has to serve over 3 million of users, so they are forced to constantly pay to keep this server running

2 completely different products

[covert lake]

elaborate in #1192011222023950368

[covert lake]

It's Weights.com, not .gg anymore

there have been changes since they have to pay for a 24/7 server that runs powerful gpusher

weight's highest plan is 24.99$

kits.ai highest plan is 59.99$ lol

[covert lake]

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @hidden grotto
• https://weights.com/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• Suggested Models for Realtime Voice Changing (Wokada)
• https://voice-models.com/
• https://thevoicemodels.com/ (for Turkish Models, login required with discord and level 2 on their server)

if there isnt one, you can:

• make it yourself with our docs guides
• Be aware that we don't allow any model request, so don't fall for any "pay me 20 dollars and i will make the model for you" dm

[hidden grotto]

:wave: @covert lake, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models

[proud raptor]

It ain’t there

Ts bot don’t understand 🤦‍♀️

[covert lake]

Weights.com has recently done some changes to the limits due to the extremely long queue crisis there has been recently, along with the expensive they have to run 24/7 with powerful gpus, AI is expensive to run

[covert lake]

search in all sites i sent, if there's none, make it urself

[covert lake]

Weights.com did those changes due to the extremely long queue crisis there has been recently, along with the expensive they have to run 24/7 with powerful gpus, AI is expensive to run

Just before people needed to wait half a day for an image gen, from what I see on their server

[pale epoch]

then limit the FREE members, not your paying customers who now get 95% less for their money now.

[terse heron]

Well, it’s as I said I’m just gonna try to be chill and be patient with it wait for it to change in the next update or eventually happen to fix a few flaws and itself for the better, maybe not today or tomorrow, or the day after, or next week, month, or year

[covert lake]

I don't work for Weights.com on their site, nor can control anything of its limits

All I can say is that they have done this due to recent costs, and that it is outlined in the Terms Of Service you legally agreed to that you agree with changes at any time

Tho it is true its been a big change, as I said again, this is the AI HUB server and I don't work for Weights' site, if you want you could give suggestions in their discord.gg/weights but the reason is due to cost

You can cancel your subscription at https://www.weights.com/settings, and ask for a refund dming Beatsboy

[covert lake]

I don't really think they would brinng as much free things as they did tomorrow, tho they are probably listening for feedback in their server discord.gg/weights

ofcourse, the reason why they did this is the cost for them, as from what I read on their server, seems like just 5% of the users pays

[terse heron]

I’ve already been in there, no need

One day they’ll change this, or won’t
I can fully understand how expensive it can be to run AI
Whatever it takes as long as some of us can eventually make our creations 5 minutes or 15, I won’t be pushy or impatient like a few members in that server protesting

[whole sun]

[urban phoenix]

Can you please tell me where to share (Ai Music - Original lyrics) ?

[covert lake]

promos ain't allowed

[urban phoenix]

Thanks for clarifying

[rough fable]

https://github.com/gasthell/continuous-thought-machines-nlp

I think it will be useful)

I made an nlp implementation of Continuous Thought Machines (model for CV). I have a lot to do to reveal the potential of this model in nlp. But for now this is it, I will release updates soon

[wraith crystal]

Hi! I am Lia, and i am looking to connect with AI agent builders / prompt engineers.
Building workflows using LLMs – happy to exchange ideas or possibly collaborate! Maybe you open to discussing?

[paper prawn]

Guys what happened to arisu voice in this channel it's gone😅?

[final spruce]

Sg ai

Ai

Where the ai at lol

?

[paper prawn]

Wrong server

[polar oriole]

3 minutes limit for audios now?? Be serious y'all

[foggy beacon]

AI

[foggy beacon]

Huh

[covert lake]

Weights.com has recently done some changes to the limits due to the extremely long queue crisis there has been recently, along with the expensive they have to run 24/7 with powerful gpus, AI is expensive to run

btw discord.gg/weights is their server

[native pecan]

hi, i'm an experienced software developer and i suck at maths, i never really meshed with maths at beyond school level, but i want to understand more about how machine learning works under the hood. where can i go that explains it like im NOT a PhD researcher?

ive read some articles and seen some youtube videos but they go too far the other way, explain it so abstract i cant see how that can be code, if you get what i mean

"heres a mesh of weights and a path through them and somewhere... magic happens, and you get a result out because maths voodoo"... too simple

when i say i suck at maths, i use only what i need, in my mind the computer is there to do the maths not the human

[chilly lake]

all machine learning is "y = mx + b"

[hallow hedge]

Hi people, I rlly want to study about how computer vision works. It would be rlly great to connect with someone who knows vision.

[native pecan]

when you explain it like that, it sounds simple

[hallow hedge]

Linear regression

[native pecan]

but im guessing its where you do y = mx + b with ten billion b and fifty billion m and x, thats the hard bit

[hallow hedge]

You don’t do it , your model does. So relax

[chilly lake]

no, you do 100 million y, then you roll a dice on each and if the result is >n you pass the value down

[native pecan]

thats the thing, i want to understand how these things work under the hood

at least the beginnings of

[hallow hedge]

Ok so until and unless you wanna build the model

[native pecan]

i want to know how the sausage is made 😂

[hallow hedge]

Building model is difficult

But like feeding dataset and then taking out output from it, is easy

I suppose

[native pecan]

i think what i'd like to do is to make more of the ecosystem 'better', as in more performant. less of it in python more of it in C++ (where i know the language well)

but to do that i have to understand it in more depth

[ripe jewel]

Oh boy, that's why I hope there can be some AI voice site alternatives sooner or later. Something without the limited options for quite a bit.

[native pecan]

ive been playing with self hosted models, and keep coming across showstopper issues (at least for me) that i imagine solutions to, e.g. things like loading an entire video in raw frames into ram, to then infer each to a model on gpu (why), or dependency hell of abandoned models in python, stuff like that. im a dev by trade and at heart, where there are problems i see solutions

[chilly lake]

wanna make big money? learn how to write CUDA kernels

[native pecan]

but my solutions would be to bring them to where i have experience, not mess with python, perhaps that makes people recoil idk

back in a bit, day job calls

[plain lynx]

May I ask if I want to get ppl to join an Ai discord server for a project how do I do that

[solar torrent]

You mean you want some people to join a Discord server? Because my brain just interpreted as "you wanna send your Discord server link to here", which you can't do here.

[plain lynx]

No why would I do that I know it isn’t allowed

More like I need some type of help from a big server

Or someone to teach me how to

[marsh canyon]

Why would they tell someone who is competing directly agaisnt them how to grow?

Seems like a conflict of interest.

[civic carbon]

Sup nerds? 👀

[autumn heath]

bro i download voice for tts but when i put it as an model it dont work like it still uses the male voice

tehjre is baldi

[marble skiff]

Today I remembered that 10+ years ago, I thought to myself "I wish Michael Jackson sang Owner of a Lonely Heart from Yes". Now I want to listen to that with AI. Can you help me? I prefer not to learn anything about AI: I just want to listen to that song.

[wicked vessel]

where i can download the Ai voice changer??

[azure valley]

hello

[unkempt glacier]

hello where i can download the Ai voice changer??

[full acorn]

I cant wait until AI learns how to mod skyrim so i can stop downloading the shitty human made ones that crash if i step on the wrong fucking twig

[flint mortar]

can i get stream permissions? i wanna stream how im working in my project 🙂

[hallow hedge]

Have you heard abt DSL ?

[violet star]

yo hello anyone got time can join me on my vc 1 please?

@signal cairn can i have a 1 minute from your time please?

@peak dome ?

[meager nexus]

I really wish there was a reasoning model (honestly a must have for me now days for llm models since it's still plenty fast for most context I would use an llm for) and one that has vision built in. Since I want to share screenshots to trouble shoot issues. That and tool use too. Qwen3 30B MoE is amazing but just wished it can see.

[plain lynx]

What do you all think about Ai making humanity not do their jobs and Ai taking our jobs cuz for me it’s like we already had this situation beforehand when making machines but then we needed jobs for Thoes machines to make them and make sure they work and this thing has happened many MANY times over same with oh no Ai will make us not know what’s true and not true on the internet we always had photo shop Wich already does that and ppl said the same thing about it we humans always find a job find a way also Ai can never be perfect due to humans brain there is a whole part of the human brain that regonizes humans and things that look like humans as scary so tbh I feel like it would change nothing just like those machines in factories just like photoshop just like computers just like cars it all had the same fear of us being replaced it’s right that Ai is really smart but it costs money a lot of money and that’s why it’s not replacing humans at job even if it replaced us in the jobs we have right now like stores and selling thing we will still find a job like working on thoes ai and llms it’s weird yk each time it happened each time it will happen the answer will always be the same we will always have jobs and that’s the hardship and sweetness of life we always will have difficulty and we will always have fun no matter what world what time we always will have it

[still sigil]

https://tenor.com/view/minor-spelling-mistake-figure-out-what-you-saidf-decrypt-what-that-means-me-trying-to-find-what-that-means-what-that-means-gif-3174979644739797084

[meager nest]

how do i setup vc client for mac?

[plain lynx]

Oh sorry mate I am not native to English so I don’t use punctuation

I’m an Arab soooo yeah sorry bout that

I also did type out words correctly I just don’t know punctuation

[polar flax]

https://tenor.com/view/wesker-albert-albert-wesker-resident-evil-resident-evil-4-gif-5143688443782660001

(unless you put punctuations ofc)

[plain lynx]

Oh wait ( short form ) : Ai can’t take human jobs we had this before with photo editing machines in factories and others

[plain lynx]

Mate for like the third time I’m not a native English speaker I’m an Arabic person

My bad the second time

More correctly an Arab

[polar flax]

we won't expect full machine autonomy

but it can still be good human-machine synergy

rn self-driving robotaxis are still rather controversial

[plain lynx]

We had same fears when starting using machines in factory’s and even older with cars and older with phones ( not the phones we have today the old calling machines ) so yeah

So I don’t get why people are like AI WILL TAKE US OVER nothing in life is easy and there won’t be anything in life that’s easy

We had it before and we have it again

Like the ones before it and after it

[polar flax]

but if machines can reach independent level of autonomy

I'd dream of machines self-sustain on the Moon, Mars, or beyond

[pure pelican]

whats the actual most efficient real-time local vrc?

[native glacier]

Didn’t read, im really happy for you though, or i feel your pain

[native glacier]

Some words are misspelled

From a quick glance

[merry stump]

how i use rvc tool

.... (╯°□°)╯︵ ┻━┻

[river kestrel]

Do a lot of AI sites use the same like provider or something? Im trying to make some AI images and im noticing that both grok, chatgpt, and an AI image remover site seem to be stuck on generating something. Is this a coincidence or no?

[still sigil]

@balmy walrus not what this server is for

[lost niche]

I created an AI Chatbot LunaTalk that can play RPG games,role-playing, funny quizzes,and more. It’s free. I think it’s pretty fun myself. If you're interested, feel free to chat and take a look.

[chilly lake]

there are plenty of proprietary image generation models

chatgpt uses o4, grok uses something made by xAI

[winter viper]

what software should I use for voice cloning RTC on win pc

[native glacier]

┬─┬ノ( º _ ºノ)

[fast basalt]

Lips Sync help needed please :
I was using LatentSync with very good results and upgraded to 1.6 but the 1.6 doesnt work anymore on my setup and I need to rollback to 1.5 but I cannot get it to work (Asking for Insightface while it is not supposed to).
I am looking for a LipSync tool using video and sound (no generating videos, image or sound, using game footage with my voice) that work well just like LatentSync (that have some Lips tweaking settings and the lips/mouths actually can move).
Anyone have something to suggest please?

[covert lake]

There are different Text To Speech (TTS) AIs:

GPT So Vits: RVC isn't as good as GPT So Vits for tts, but gpt so vits (few shot tts, which means needs just a lil training for models) can't use rvc models (and viceversa), and its only limited to: english, chinese, Cantonese, japanese & korean, if you wanna check gpt so vits instead, read https://docs.aihub.gg/tts/gpt-sovits/

Freemium 11labs: Easy way to do TTS is https://elevenlabs.io/, you can't use RVC model on this but its a mostly premium easy way for good quality TTS

FishSpeech: FishSpeech is a 0 shot (no explicit training needed) TTS, if you got a good pc you can use it locally else use their site

You can check TTS in our tts index

With RVC Models:

RVC is natively for Speech To Speech, but forks such as Applio have built in tts (using Microsoft Edge TTS to make a generated tts audio, which i suggest you to choose a tts model that is the same gender and language of the rvc model you wanna use, and then convert it with rvc)

If you wanna do tts locally with RVC Voice Models (if you got a good pc):

• You can get Applio in our docs

If you don't got a good pc you can do tts with RVC Voice Models on cloud:

• Ilaria RVC Zero (Running on A100 GPU, free fasted rvc on cloud) and the guide

• Use Applio UI Colab (with google colab T4 free daily limit gpu)

• You could try another tts from our tts index and use the output as an input in rvc

please use #1192011222023950368

[covert lake]

elaborate in #1192011222023950368

[covert lake]

elaborate in #1192011222023950368

[covert lake]

don't randomly ping admins

[covert lake]

elaborate in #1192011222023950368

[covert lake]

elaborate in #1192011222023950368

[covert lake]

elaborate in #1192011222023950368

[covert lake]

elaborate in #1192011222023950368

[covert lake]

try asking in #1192011222023950368

elaborate in #1192011222023950368

[sudden lava]

wahts the best ai rag?

[still sigil]

@velvet sonnet no advertising

[elder willow]

Is ChatGPT good at Thai

[ripe adder]

hey there everyone

[lament charm]

whats up gng

I love the voice models from here

Do yall know if characters can be face modeled too? like "v tuber" but I rip a character's face from my favorite game instead of just their voice

I imagine some degree of deepfake

[quasi dome]

Zerogpu error

How do I fix

[covert lake]

please elaborate in #1192011222023950368

[rich thorn]

juice wrldfan

lol

[elder willow]

Can I use the AI chatbot ChatGPT for the Thai language

[chilly lake]

why don't you just try?

[hidden talon]

help is there anyone with an ai humanizer?

[chilly lake]

if you mean AI text to something that sounds human, it is just another AI text gen

sometimes it works with just a proper prompt

For my test, I generated a “What is Data Recovery?” article using ChatGPT 4.1 Mini. Then I ran it through GPTZero, ZeroGPT, and Quillbot’s AI checkers - got hit with 90-100% AI detection. Not great! But this is what I need!

Now, let’s see if I can actually humanize this AI content and fool the checkers. What if it’s actually that simple - just type in a prompt and fool all the detectors? Let’s put that to the test and try every trick out there, both one by one and all together:

Write like you’re chatting with a friend:

    Vary sentence lengths - mix super short (3–5 words) with longer ones (20–30 words).
    Toss in personal asides (“I think,” “honestly,” “in my experience”) and occasional “And,” “But” or fragments.
    Use conversational flair: “you know?,” “well, actually…,” light sarcasm or humor.

Keep it fresh & human:

    Swap out stale phrases (“very important”) for punchier words (“crucial,” “game-changing”).
    Break a few grammar rules naturally - end sentences with prepositions, split infinitives.
    Sprinkle in real-life examples or anecdotes with exact numbers or pop-culture nods.

Structure & tone:

    Mix paragraph lengths - one-line punches and longer blocks.
    Prefer active voice, simple ideas, no corporate buzzwords or clichés.
    Use creative transitions (“anyway,” “that said,” “still”) but don’t over-explain connections.

Stylistic touches:

    Emojis are out - use ellipses (…), em-dashes - sparingly for rhythm.
    Hedge when it feels natural: “probably,” “maybe,” “kind of.”
    Add tiny detours or soft self-corrections (“actually, let me rephrase…”).

The goal: Write like you're texting a smart friend, not submitting a report. Choose words you'd actually say out loud.```

[severe spindle]

i know theres miku rvc modles but wheres the teto ones?

[regal flower]

whats the best software that i can use to train a voice

[quasi glade]

Can someone point me to the right direction to request a AI model?

[covert lake]

None, we don't allow requesting models

[covert lake]

Elaborate your hardware in #1192011222023950368

[covert lake]

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @hidden grotto
• https://weights.com/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• Suggested Models for Realtime Voice Changing (Wokada)
• https://voice-models.com/
• https://thevoicemodels.com/ (for Turkish Models, login required with discord and level 2 on their server)

if there isnt one, you can:

• make it yourself with our docs guides
• Be aware that we don't allow any model request, so don't fall for any "pay me 20 dollars and i will make the model for you" dm

[hidden grotto]

:wave: @covert lake, How can I help?

Available Commands:
• @weights find <query> or /find <query> - Search for RVC Voice Models

[quasi glade]

Thank you very much appreciate the information

[elder willow]

What does fidelity mean

[chilly lake]

The degree to which an electronic system accurately reproduces the sound or image of its input signal.

[olive crescent]

how can i create videos ai with polish sound

[devout matrix]

Well if your referring to making Ai music you can definitely use SUNO. It's a very good ai music tool you can use to make your own music. I'm a user of it as well, but if you want to get more extreme tools in it you'll have to pay.

But other than that, I'm sorry 😫

With what?

I'm not sure how I can help u with that

I'm only used to making Ai music that's all

[loud wraith]

How i can train ai vocies when no premium

[rich basalt]

https://tenor.com/view/cat-arab-helicopter-gif-24466643

[rich basalt]

There's also Udio, but not there yet in terms of performance accuracy

[rich basalt]

Or Feedback Loop

[regal flower]

Use applio

U have to have a good gpu

In general a good computer

[tired flume]

arabic is arab in english

[rugged ravine]

does anyone here have experience with ai generated music?

[loud wraith]

where i can do get that ?

[regal flower]

its on the rvc guide

https://docs.aihub.gg/rvc/local/applio/

[polar flax]

other than Weights site, I don't recommend any paid services that may be sketchy

or you could consider these alternatives

Local (8+ GB RTX gpu recommended): https://docs.aihub.gg/rvc/local/applio/
Cloud:

• Applio colab: https://docs.aihub.gg/rvc/cloud/applio-colab/
• Applio kaggle: https://docs.aihub.gg/rvc/cloud/applio-kaggle/

[lucid ibex]

Is mj still in here?

[still sigil]

no he died in 2009

[lucid ibex]

My bad

I meant [Not] MJ

[weary stratus]

Where can i read w okada voice changer update log

[lucid ibex]

Idk

[harsh junco]

most likely in the github page

yup, the changelogs are in the read-me's (english ver: https://github.com/w-okada/voice-changer/blob/master/README_en.md )

[lucid ibex]

Alr I'll see you all in a few years

[pulsar rover]

yo