🧬│ai-chat

[gentle trench]

I think i'd get deleted

[night geode]

you make up about half of the sum total of messages that happen in this server

they wouldnt ban you

[cunning dagger]

@topaz needle

[topaz needle]

wsp

[wanton citrus]

im already a girl

[gentle trench]

eh

just look here https://discord.com/channels/1159260121998827560/1175430844685484042

[uneven quiver]

Hey guys, what's new for cloning an AI voice? The last time I tried it was with Applio 2024... there must be newer, higher-quality programs now?

[wanton grove]

Translate!

Click here to go to the translation link where you can help us translate me into multiple languages. If you know any languages other than English and have time to help in translation, please go to that link and start translating!

If your language isn't there, let us know in the support server and we'll add it so you can translate!

[night geode]

pretty sure Applio is still the go to for voice cloning rn

[uneven quiver]

Thanks for your reply, but I get the impression that nothing has changed since 2024, still the same greedy process, nothing has been simplified or improved?

[night geode]

only been about a year, 2026 is still early. and theres not much profit in voice cloning so progress will be slow for updates

[uneven quiver]

Is there a quick guide for cloning as perfectly as possible? I know very well how to extract voices, but cloning them is another story; I always end up with defective models despite using high-quality assets

[gentle trench]

try the legacy core v1.5 32k pretrain

the pretrain can be found here https://discord.com/channels/1159260121998827560/1453062294446805244

[uneven quiver]

Thanks brother, but I don't quite understand the concept. Is it about using a pre-trained model to perfect a voice with my assets?

[gentle trench]

it would be a change from using the original pretrain to something newer to see if that can fix the issues you may run into

[uneven quiver]

I need to see someone practice; I don't really know how to use it without making a disaster 🤣

[small herald]

@spare current is randomly adding people possible spam

[gentle trench]

probably yea

[small herald]

And they got hella roles

[grave flower]

Hi everyone, could you help me download the program to create voices for my characters, please?

[solar torrent]

For help about Applio RVC, ask in #✨│ai-help or type "-rvc" in #🤖│bots.

[noble halo]

Hey does anyone know a good vocal isolation tool? CapCut makes you pay. I do have iZotope RX 11 Audio editor but wondered if there was anything that works better. I have a long YouTube video that I need to use it on.

[odd knot]

mvsep

mel roformer deux

[noble halo]

Is that something I can get for free?

How long does it take?

[odd knot]

mmm yeah me thinks

or you can run the model locally with uvr5 if your pc's powerful enough

the only downside to mvsep tho is the queue

[noble halo]

Oh.

I don’t wanna deal with a queue.

[noble halo]

What does CapCut use? Can I get that locally for free

[odd knot]

then try uvr5

or create a new account on mvsep

[solar torrent]

There's UVR5.

[noble halo]

I have a MacBook Pro 16gb ram

[solar torrent]

Apple Silicon or Intel Mac?

[noble halo]

Silicon

My video is about 45 minutes

[solar torrent]

There's UVR5 compiled for Apple Silicon Mac's GPU as well. https://github.com/Anjok07/ultimatevocalremovergui/releases/download/v5.6/Ultimate_Vocal_Remover_v5_6_Roformer_beta_0115_MacOS_arm64_hf.dmg For more help about UVR5, better go to #✨│ai-help or #1192011222023950368.

[noble halo]

What’s the link for windows?

[solar torrent]

[clever sleet]

damn bruh why my audio be so choppy

[gilded raven]

i need help by someone when i downloada voice model its just a .flac or .wav and i cant put it in the voice changer

[odd knot]

oh brother

https://tenor.com/view/daitaku-helios-umamusume-67-brainrot-gif-4785550353554980182

[half halo]

where's i can get the ai hub app guys? purchase or what?

[still sigil]

there is no app

[solar torrent]

https://tenor.com/view/sixseven-six-seven-six-seve-67-gif-14143337669032958349

[solar torrent]

There's no program or app named AI Hub or AI Hub by Weights. What you use to go to one server and ask a query is "Discord", which is the platform application itself.

[hardy acorn]

Yo I have a freakin question..

Does The voice changer work on 4 gb ram pc with 2 gb ram graphics card

[solar torrent]

No.

[hardy acorn]

Alright..

Thanks

[languid oasis]

rvc v3 whenn

[odd knot]

never

atp anyone here could make rvc v3

[solar torrent]

Anyone can claim they made RVC v3 real, because generally the most recent and available RVC version is v2.

[proper kestrel]

👋

[tropic oracle]

👋

[half halo]

So, how to use the voice models? Is there a tutorial?

[solar torrent]

That's an expected response. For help about a voice changer, better continue in #✨│ai-help.

[safe umbra]

Hmm

[left sphinx]

AI for moderation seems like a really good use case

almost instant decision making if you feed it a proper rule set

and 0 bias or frustration

i just implemented it for image uploads and it works really well

[sharp forge]

Hey everyone 👋
I’m starting a small build-in-public experiment today.
The goal is simple:
show up daily, build something useful for founders/students, and share what I learn along the way.
I’m keeping expectations low and curiosity high.
If anyone here has done something similar, I’m happy to learn from your experience.
Day 1 starts today.

[true osprey]

New release - https://www.youtube.com/watch?v=_-1-xolrxEE

[icy maple]

Hi everyone! I'm looking for a mature female RVC model with a Spanish (Latin American/Argentine) accent. I need a voice for a 'villain' character, so ideally someone with a deeper, seasoned tone (50+ years old). Does anyone have a recommendation or a link? Thanks!

[left sphinx]

https://discord.com/channels/1159260121998827560/1454635655975145544 this is what i'm working on

adding a poll system to the post system

[gentle trench]

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

[minor blade]

Look on the #1175430844685484042 channel altho i highly doubt there's any model like that. So in that case make a request on the #1159289738314919936 channel

[dusky thistle]

Hi

[meager ocean]

Hey 👋

[dusky thistle]

@meager ocean I play with ai from time to time

[meager ocean]

That's good 😊

Where you from?

[dusky thistle]

The us

I think ai is funny

[meager ocean]

Wow, I'm from Singapore

[meager ocean]

Why say so?

[dusky thistle]

Cause it's funny making eldritch abomination of Pinkie Pie with huge teeth 😂

I'll assume I can't post anything until I lvl up, right?

[meager ocean]

Nah, I guess you don't really know much about AI modeling then....
If you're talking about dem stuff here then I'll tell you it's nothing compared to an AI character model...
The girl on my profile is an AI not a real girl though

[meager ocean]

Don't really know though, I'm not always active here

[dusky thistle]

I only know civitai, mid journey, or bing

[jade thorn]

what do i need to do when it doesnt make any sound

[tropic oracle]

👋

[proven onyx]

hat einer gleggmire voice?

ja brauche auch mein akh

[gentle trench]

I don't like where this is going

[woeful bolt]

hey who makes images with sora here?

[gentle trench]

Doesn't sora just make videos?

[woeful bolt]

it also makes images

[gentle trench]

Oh I didn't know that 👍

[woeful bolt]

their images don't have a watermark

[dreamy bloom]

yoo

can i ask someone a question

does anyone know a ai thats unsensored, im trynna add a video and edit it a bit

[daring epoch]

hey

[proven onyx]

yo whats the best girl voice

[gentle trench]

Why :(

[digital narwhal]

gng how do u make it not echo

can smb help

[unreal tree]

Who knows how to create Ai adverts

[thin apex]

Hello 👋
Yes, I can help, but I’ll need a bit more detail because “RVC error 2” can have different causes.

Please tell me

Hello, I am trying to create voice models, but RVC v2 keeps giving me an error. I am doing everything correctly, but I still cannot create the voice models I want. Could you please help me

[gentle trench]

You should try using applio

-rvc

[pine acorn]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[lethal lark]

Anyone know how to fix this error?

pkg_ressources is deprecated as an API.

[gentle trench]

There is a guide here of how to use it, if you followed anything from YouTube it's outdated @thin apex

[thin apex]

Yes, I watched it on YouTube; the videos are 2 and 3 years old. I really need it now, but it’s not working and gives an error @gentle trench

[gentle trench]

Should ask in https://discord.com/channels/1159260121998827560/1175430844685484042 for someone who knows how to use applio locally as I don't know how to use it on local

[noble halo]

is uvr5 local on windows not have a quque?

i have like a 45minute video I need to vocal isolate

is there an easy vid guide on how to install

i have a 5070ti gpu

[limpid cave]

does anyone here use Gizmo AI?

[lethal lark]

PyInstaller\loader\pyimod02_importers.py:378: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.

[lethal lark]

@gentle trench

[gentle trench]

I don't know how to fix errors sorry

[noble halo]

Which model in local uvr 5 has vocal isolation tool

[gentle trench]

if it doesn't say it's for reverb, echo, noise, ect it's probably a vocal isolation tool

[noble halo]

Trying to take a mp4 video that’s around 45 minutes and isolate the vocals

It just says this

[gentle trench]

o

I don't use the local version so idk lol

[noble halo]

Oh 😭

I have no idea how to use it the segment size should probably be small if I have a 45 minute video

[odd knot]

try converting a video as wav/flac and select that audio file as input and the folder you wanna put the isolated vocal in as the output

as for the model itself for isolating vocals i'd recommend deux for now

it should be available here https://huggingface.co/becruily/mel-band-roformer-deux/tree/main

[eternal geyser]

Hi everyone! Excuse the interruption; I’m excited to share an entirely open-source research project I've been developing: Manifold Laboratory.

We are proposing a shift from purely statistical heuristics toward a physics-informed architecture. Instead of standard attention, we use Riemannian Geometry and Symplectic Integrators to model "thought" as a particle trajectory on a curved manifold.

Key Technical Highlights:

• Complexity Breakthrough: We optimized the manifold state updates, reducing the typical O(d^3) cost of metric-tensor operations down to O(d) using a custom Symplectic Heun Solver.
• True O(1) Memory Scaling: Unlike Transformers (O(N^2)), our architecture maintains constant memory relative to sequence length.
• The "Parity Rumble" Benchmark: We achieved 100% accuracy on a 10,000 token context length—extrapolating from a training length of only 20—using just 581MB of VRAM on a consumer GTX 1650 (FP32 precision).
• Active Inference: Built-in Plasticity, Singularities, and Dynamic Time (dt).

No tricks, no quantization—just pure differential geometry and CUDA optimization. I’d love to hear your thoughts or see some of you break it!

🔗 Repo: https://github.com/Manifold-Laboratory/manifold.git

📊 Benchmark Data: https://raw.githubusercontent.com/janxhg/MANIFOLD/main/tests/benchmarks/results/gfn_superiority/parity_generalization.png

Apologies for the intrusion. I simply wanted to share my fully open-source project, which I believe represents a significant leap forward in how we conceptualize AI architectures.

[dark rover]

some of the latest series nvidia cards have already reached a price of 6700+ euroes a piece in some places. You might want to make things work on AMD cards, which too have options for high vram.
rocm has been made windows friendly if that is the problem, and there is HIP to help convert cuda instructions to rocm.
and as a fallback there is also vulcan or dml if really needed. Keep in mind that aside from the initial cost, nvidia gpus cost a ton of electricity to run, so you do need a budget for that.
considering the extreme price hikes, I don't think the future lies with nvidia datacenters. Heck even openai doesn't believe that. A while back they went for AMD instead due to less power usage.
It's silly to focus on CUDA, is what I think basically. You're locking yourself into an ecosystem designed to extract money; not for developmental progression.

https://openai.com/index/openai-amd-strategic-partnership/

[eternal geyser]

Thank you so much for this feedback. Honestly, this is a perspective I hadn't fully weighed yet, and I really appreciate you pointing it out.

As I’ve been developing Manifold on a modest GTX 1650, focusing on CUDA was my 'path of least resistance' to get the initial proof-of-concept running. However, your point about the NVIDIA monopoly and the shift toward AMD/ROCm for sustainability and accessibility is truly valid.

The idea that focusing strictly on CUDA might hinder 'developmental progression' really hits home, especially since the core mission of this project is extreme efficiency and democratization. I will absolutely keep ROCm and Vulkan in mind as I plan the next stages of the roadmap. Thank you for the wake-up call and for the technical leads (HIPS/Vulkan/DML)—it’s exactly the kind of insight I was hoping for by sharing this.

[gentle trench]

I belive here is a better space to share your creations ^^
https://discord.com/channels/1159260121998827560/1359898289335566570

[eternal geyser]

sorry, but I can't post there, it won't let me

[gentle trench]

hmm that's odd

maybe one of the roles you did not grab will allow you to post there

in here

[eternal geyser]

Maybe that's it, I'll check it out

It still won't let me post, even though I have all the roles

[gentle trench]

maybe you need to level up then, that's all I can think of

[eternal geyser]

It's likely

[hollow storm]

Reminder that the ai bubble will soon pop. and note that I only use this server for voice acting and nothing else

[gentle trench]

Most people use this server for the voice models, that's it

Only good use of ai in my opinion

[hollow storm]

true

[oak spire]

smp ip : gravehold.nethr.nl

[solar torrent]

Don't dox someone's address.

[compact spire]

Goodness I wish i had nitro

[bold lily]

ok

nb cares

[hearty fractal]

How to get svc Collab link with ui

[solar torrent]

Still using so-vits-svc in the big 26? That's crazy.

[hearty fractal]

So what to use for cloning voice?

[split sand]

hi what is the most realistic voice changer actually and the fastest with a ryzen 7 3800x + gtx 980ti?

[solar torrent]

Applio RVC is superior.

[hearty fractal]

@solar torrent do you have any colab link for cover making on applio

[solar torrent]

Type "-colab" in #🤖│bots.

[solar torrent]

https://tenor.com/view/i-saw-what-you-deleted-cat-gif-25407007

[humble stag]

hi anyone wanna collab

[left sphinx]

yea want to help me test my platform?

[humble stag]

DM and we can talk more bout details personally @left sphinx

[left sphinx]

i DM'd, thanks for wanting to help

[round pine]

HI WHAT IS THE BEST VOICE FOR GIRL VOICE

[bold lily]

larp

[solar torrent]

Kasane Teto

[round pine]

ok

[gentle trench]

Agreed

Use my model

[unique rivet]

👋

[little shard]

👋

[gentle trench]

<@&1159293140440723499> cook the duck

[full wind]

can we make ai corn here ?

[gentle trench]

Did you read the rules :(

[full wind]

i can't read

can we >

[gentle trench]

I'm sorry for your loss

[full wind]

EFM>EFJ

[full wind]

wdym

give me a server

[low basalt]

This is a 54 minutes Epic fantasy Ai movie ive been working on since august.. here is the link if you want to watch it: https://youtu.be/6l4w_utNdj0

[jagged gorge]

So i tried my first own training a model with the help of geminin pro (yeah i know it does mistakes or tells you things not accurate but mostly it is accurate enough for me to find the answer i need).

My workflow was raw data > UVR5 > RX 10 (not sure if you can get RX11 "cheap" without bad things coming with it) > applio.

I have a 4090 and my results... are not great. Now i'm a bit of a search of the root problem.
For example gemini insisted on 48k sample rate for best "quality" even if the audio itself not even reaches that or how it explained it as a "container".
And yes when it comes to training your own model i did read the guide from applio sites and tried to use the right settings but i have a feeling i did do set something wrong.

Some settings are at least for me a bit confusing.
Batch size for example it.
Lower takes longer but is more accurate but bigger trains faster but overtraining happens faster?

When i have like 1h 48khz that is very diverse (many emotional words or noises) of dataset and used Titan pretrain. I feel like i did something wrong with the other settings even when i tried to go like the applio guides say.

There is for example a "Silent training files" part and i'm like.... when do i know i have enough pure silence? When i have 1h+ audio files and was manual cutting many things out and shorten many breaks between words (not to harsh but sometimes there was still music or other noises in the background).

So yeah if anyone with experience could help me out i would be very thankful. I kind of feel like i did some small mistake that lead to big mistake in the training.

[sudden coral]

40k is perfectly fine, you don’t need to force 48k. quality will be great as long as your dataset is

I’ve found batch size between 5-8 to be the perfect balance of accuracy and capturing essential characteristic details while still learning larger patterns

silence: if you have roughly 0.5s-1s between phrases you should be fine, don’t overstress it. if you’re training in Applio, sticking with the default of 2 silent training files should be perfect

[jagged gorge]

So gemini tries to steer me to things like "you have a 4090 so go with batch size 16 because you have the gpu for it" and the training is a lot faster. But does it kill the quality?

Yes the silent part is very confusing because what is when there is not much of "pure" silence? What is if there is to much?
Can i always stay with the standart setting of 2 and not being total confused if i have to much or less of it?

So i use Spek to check the audio files and what there real peak of sample rate is. And most are around 16k x 2 = 32k (youtube ripped with yt-dlp) and some can be around 20-24k = 40k-48k. So it is a bit mixed.
So i really try to learn everything correctly but there is much that still confuses me to do it right.
And i not trust gemini to be total accurate because alone when it comes to menu settings in UVR5 or RX 10 it is good in telling the models or modules to use but seems to be often confused about settings. So i often correct it to then give better response.

[sudden coral]

well here is your human help 😁

batch size 16 imo is crazy big, it most definitely affects model quality and will overlook finer details and nuances, but if you’re not concerned about that then go ahead. the voice will just sound overall “smoother” without those details.

just trim your long silences to about 0.5s-1s, keep silent training files to 2, and you should be good.

[hollow wharf]

GUYS DOES ANYBODY KNOW A GOOD GUY MODEL

[pine quartz]

Just wanted to ask because i have no clue. Does Epochs matter for the models? like, more is better or lower? cause i didn't really deep down to modelling or something familliar with it.

[misty frigate]

more epochs is better until your model overtrains which is a problem I haven't had recently

[thin apex]

If I send you a vocal in a WAV file, can you prepare a modeled version of that vocal for me?

[still sigil]

@snow wing no begging for money

[grim phoenix]

help me on why the voice changer is cutting out and how to make it not be delayed

[still sigil]

there is no beta

[cinder steeple]

does anybody know the best AI model to use for code generation through api key with azure?

[left sphinx]

why are you using azure

[cinder steeple]

api key

its cheaper than openai and others

[left sphinx]

interesting i don't use it

and i won't be telling anyone what i use for free access to Opus 4.5

^ best emoji on the server

[cinder steeple]

pls do

i need for a competition

[left sphinx]

sry can't it's secret

[cinder steeple]

😭

😭

[cinder steeple]

i wont tell 👀

[left sphinx]

that's what she said

[cinder steeple]

calm down yo 😭

[left sphinx]

https://tenor.com/view/ugly-plankton-meme-ugly-plankton-ugly-face-plankton-meme-plankton-gif-15836444981074643801

[cinder steeple]

😿

[solar torrent]

Minecraft.

[tropic oracle]

👋

[little shard]

👋

[unique rivet]

👋

[cinder steeple]

👋

[cinder steeple]

i mean for web apps.....

[left sphinx]

[jagged gorge]

Did anyone hear about "MaskVCT" or "StreamVC" ?

[swift sun]

Anyone here do commissions.

[still sigil]

commissions are not allowed

[swift sun]

Ok. Because I saw that was updated in the server rules about a year ago and still saw people saying that they do free commissions.

[still sigil]

free is allowed, just not paid

#1159289738314919936

[swift sun]

Yeah I meant free commissions. Thx

[ornate pilot]

Is there French egirl voice ?

[fierce hornet]

can someone help me it is giving me problems

[grave vigil]

Hi, anyone knows whos voice was used for this model?https://huggingface.co/ZokaxDesu/e-girl/resolve/main/e-girl.zip?download=true

this is the model from void

[wooden snow]

does anyone know how to make this transition from 0:04 to 0:08, from one scene to another, i used "camera pans right" but each time it looks crazy, any tips? https://www.youtube.com/shorts/HLS9u6JGaXg using kling ai

[odd knot]

<@&1159293140440723499> i think he got hacked

[civic kraken]

i people, not sure where to start but i would like to build a website AI generator for webdesign and hosting for small businesses but dont know how to go about it. Can anyone help?

[potent sparrow]

how to get more video cradit on sora 2 free

[hot bridge]

guys one question.how active is this channel

[dawn temple]

Not active unfortunately

[left sphinx]

did anyone want to try this?

if your bored and want to share your AI generations

[merry sage]

What would you guy say is the best Dataset to fine-tune a LM that can process images for reading intersocial cues, body language and fascial expressions in a nuanced way. I would probably be using Llama 3.2 Vision and fine tuning it in Google Collabs. I would run the LM locally in LM studio

[gentle trench]

<@&1159293140440723499> blow them up

[cold crag]

https://engage.mygamesonline.org/dragon/

made with ai

[swift sun]

What is the epoch range mean for voice models

[gentle trench]

that's just how many cycles of training the model went through before sounding good to whoever made it

[static delta]

Hi

[low basalt]

From USA you%

you?

[gentle trench]

shit

copied the wrong thing

free kills for moderator

I hope I didn't get hacked copying that

[fading garden]

banning them for getting hacked instead of waiting for them to recover their account and join back

[still sigil]

thats what modmail is for

[fading garden]

u guys got modmail

[still sigil]

then they recover their account, they can dm modmail and we will unban

[fading garden]

?

[still sigil]

yes

[fading garden]

then i take back what i said

https://tenor.com/view/happy-happy-dog-dog-happiest-dog-super-happy-gif-17885812

[still sigil]

@serene sluice

[fading garden]

sussyboy if ur still here what version to get for windows 2.1.4 alpha or 2.0.78 beta

oh wiat

wait im dumb

[teal wadi]

what

why all voices from channel with voice models doesn’t work😭

[bitter shard]

How can i create my own rvc model ?

[solar torrent]

For help about W-okada voice changer, go to #✨│ai-help.

[uneven pewter]

train it

[uneven pewter]

they do

[bitter shard]

How

[uneven pewter]

search on yt

[clear lily]

👋

[versed lotus]

what is the best AI tool for frame to frame video transistion?

i have a shot thats standing far away from the subject (portrait)

then the next shot is close up

so i would take the end frame of shot 1 then pair it with the first frame of shot 2

kind of thing

where do i go for this? ideally one that allows to trial it before buying membership?

[noble halo]

anyone know any good vocal isolation tools like capcut? its nice to just split my audio if it has to be under 15 minutes and apply the vocal isolation and it doesn't take a super long time. Any alternatives? I'm not a fan of uvr5 not to mention its poor quality and wait time locally

[radiant night]

How’s it going today?

[little shard]

👋

[deep minnow]

Audacity or Reaper are my goto's for editing, not sure if that helps

[noble halo]

Do you prefer one over the other

[deep minnow]

Audacity is entirely free, reaper costs a few bucks but it's better

imho

I still use audacity a lot because its simpler I guess and I can run it on all my machines

[gentle trench]

<@&1159293140440723499> kill the hacked ones

[deep minnow]

ultimately a lot of features in apps these days, I see the trend, its to just keep shoveling features into every single app and upping the potential but adding complexity to the point of silliness. Theres always a balance to strike and when it comes to apps for audio editing I stop at reaper, you can go crazy down the rabbit hole with audio editors. I run what I consider a quasi-pro audio setup here - electrovoices not neumann and ART voice channel nothing exotic

[gentle trench]

I personally use fl studio

[deep minnow]

If it works it works, there are tons of options today. I run linux on my machines so it kind of limits options. But there are some good serious production tools, although I consider them (and not in a biased way) inferior to whats on offer for windows/osx

FL Studio looks good

[manic hornet]

Is it free

[gentle trench]

applio is free

-rvc

[pine acorn]

📚 RVC Documentations

AI HUB Docs

https://docs.aihub.gg

🍏 Applio Docs

https://docs.applio.org

[manic hornet]

I wanna make my voice

[gentle trench]

use applio to train it, all u need to do is get the audio and make sure there is no background music, reverb, echo, or other people talking

[manic hornet]

How much audio

[gentle trench]

anything over an hour isn't needed, but around maybe 15-30 minutes is good

[manic hornet]

Dang

[gentle trench]

u can make a good model from even 30 seconds if you know what you're doing

[manic hornet]

Can you do it please

[gentle trench]

I don't take commissions sorry

[manic hornet]

How

[gentle trench]

you collect the audio, or if it's your voice just record it using something with a good microphone, clean the audio using mvsep or UVR5, I don't really feel like explaining how to train a model as it's somewhat confusing but if u want more help with it I'd ask a helper in https://discord.com/channels/1159260121998827560/1159290139609137264

[manic hornet]

I really need help someone dm

[manic hornet]

Wdym

[gentle trench]

I don't make models for ppl, free or paid

[manic hornet]

Why

[gentle trench]

because I have had people in the past be kinda annoying and impatient

[manic hornet]

That's not me

[gentle trench]

still I don't want to have randoms from here dm me without even asking saying "can u make this for me" because I did smth for you

[manic hornet]

I am chill

[gentle trench]

I mean other ppl will dm me asking if I did something for you

they're not all gonna be chill

[manic hornet]

We can delete the message

[gentle trench]

https://tenor.com/view/ena-power-of-potluck-confused-disgust-stare-gif-7242501587835519731

[manic hornet]

See 😮‍💨

[fair sparrow]

does anyone know how to get vonovox to work with voicemeeter potato?
still trying to figure out that software

[compact spire]

it still can't even build an app

and no, webapps are not apps

[ripe gale]

I dont really know if this is the right place to ask; but I just made a PTH file using RVC for my voice— I did 800 epoch, and tested each and every 50 epoch saves. I have a lot of voice data; But they all sound a little too noisy (metal noise coming from AI)

Do you guys know any DAW to cover this or any ways to not.. 'completely remove' but to at least lower the amount of noise?

[gentle trench]

800? wayyyy too high

[ripe gale]

I see.

I tried 350 one, but still too noisy

[solar torrent]

Around 250 - 300 epochs are the common minimum epoch count for RVC voice model training.

[ripe gale]

I feel like
0~100 = exclude this
150~300 = high notes dont hit well but less noisy
350~500 = high notes hit perfectly, yet noisy af

Ykwim?

Do u guys not use RVC anymore to train your models?

[gentle trench]

could possibly be you need to clean the data more

[solar torrent]

RVC stands for "retrieval-based voice conversion". I use Applio RVC.

[ripe gale]

Sorry i meant the RVC project.

Appolio.. I have that installed.

But Ive never used it before so Im not totally sure of every setting

So I havent touched it yet

[solar torrent]

Applio RVC is the more recent RVC fork that adds more features to the original "mainline" RVC.

[ripe gale]

I see.

[solar torrent]

If you haven't yet tried Applio RVC, the latest version of it is v.3.6.0.

[ripe gale]

Doesnt it auto-update?

Hmm let me check the version

[solar torrent]

It doesn't.

[ripe gale]

k just updated

[ripe gale]

Would you mind answering some questions, ofc after I try it myself??

-# cus I remember learning applio but I dont really remember every differences from the mainline rvc kekw

[solar torrent]

[ripe gale]

Looking at documents on the official website >< gimme like 5

[misty frigate]

I found something kinda weird

[ripe gale]

Is the G and D models any different from what I used to use in mainline rvc? Or is there a BIS "international" G & D models out somewhere I can download separately?

[misty frigate]

in the contentvec config file, if you set apply_spec_augment to false and []_dropout to 0.0, it makes the outputted model less noisy.

Now I don't have any audio perms since i very rarely speak in this server, but the difference is pretty drastic.

[solar torrent]

That's a new information. But, is there any trade-off if you listen to the voice model audio closely?

[misty frigate]

Maybe slightly less stability with the augmentation off (but that could be because of training for less epochs)

That goes to say, I've been able to get satisfactory results with less epochs.

[ripe gale]

Ive used G&D of the KLM9T for my Korean models. And I feel like sometimes the pronunciation is 'off', so then I decided to use the default model

[misty frigate]

The models seem to do better in converting sounds outside of their dataset

[solar torrent]

While I only know about basic and more accessible settings in RVC, I haven't trained anything for a very long time, for how to train a voice model perfectly, you can try ask fellow model makers here.

[ripe gale]

Also, testing this out.

Ill see where it goes

[gray mirage]

Hey i finally found an old ai voicechanger that I used to use but its denying me access to upload my voice models into it????

does anyone have input

[gentle trench]

well it's old

time to use something new

[left sphinx]

yea they are

[lime raven]

i generated an image of a spectrogram of the vowels a e i o u with nano banana pro

i dont think theyre correct lol

@turbid trout

[coarse hedge]

Is there AI that can complete coursework for school actually in PDF form?

[uneven pewter]

yes its

[manic hornet]

Can you dm steps

[cursive crag]

hello

ive been here since 2023, has anytthing changed? regarding RVC

[gentle trench]

new voice changers like Vonovox and wokada tg fork, and much better and usable pretrains like legacy core 1.5 have been made

besides that nothin real new

[cursive crag]

is snoop dog still around?

and an italian girl named aria or something

[gentle trench]

not sure who those users are tbh

but most likely

[cursive crag]

whats vonovox? is it better than rvc?

im still using applio rvc

[gentle trench]

vonovox is one of the realtime voice changers it's not to create voice models

applio is still used lol

[cursive crag]

im tellinn u i was in the first ai hub server before it got down lol

[gentle trench]

me too

it was so different

[cursive crag]

ok so for singing models

what do you suggest?

i want to train in greek language tho not english

[gentle trench]

like for making them?

[cursive crag]

yeah like an artist or something

ive sent you a friendd request btw

[gentle trench]

just find them on yt and isolate the vocals

idk any particular ppls

[gentle trench]

u seem cool I'll accept

[snow bronze]

AI / ML Engineer (chatbots, automation, RAG)

I build AI systems that people actually use.
Mostly working on chatbots, LLM/RAG pipelines, and automation, and plugging them into real products.

Project types

▸ Chatbots & AI assistants (LLM-powered, RAG, internal & customer-facing)
▸ LLM & RAG systems (search, Q&A, knowledge tools)
▸ Workflow automation (Slack, Notion, internal APIs)
▸ AI moderation & detection (text, image, content)
▸ Image AI pipelines (tagging, filtering, e-commerce)
▸ Voice AI systems (ASR, TTS, assistants)
▸ Intelligent bots (trading, Discord, platform bots)

How I work
▸ Python + FastAPI for inference & services
▸ Embeddings, vector search, async pipelines
▸ Cloud deployment (AWS, Docker, CI/CD)
▸ Integrated into real web & mobile products

Open to projects or collaboration - DMs welcome 👍

[dusty atlas]

guys Weights is shutting down at the end of March

[lost cape]

How can I get permission to post on project showcase?

[limber nexus]

im pretty sure its only the DreamTavern product

[balmy marsh]

Hi, can you use the voices from that Weights website, and use them in wokada voice changer jsut for regular speaking? I noticed each voice I use from the Weights website, the voices always sound robotic when I speak.

[velvet crest]

WHY IS DREAMTAVERN SHUTTING DOWN AS WELL?!? I LOVED USING THE CHARACTER CHAT MODELS! 😭

[limber nexus]

Don't ask me

i only know of DreamTavern for the server

[velvet crest]

Oh... But still... I loved using the Character Chat models on DreamTavern....

I made some DreamTavern Character Chat models, and I absolutely loved using them...

[limber nexus]

i sorry for your loss

[velvet crest]

What am I gonna do. It's AI anyway. Real people who are good hearted and real characters can't be topped in my opinion anyway.

[limber nexus]

That's a good view

[velvet crest]

I mean, it's the truth. Real people who are kind, sweet, and loving, and real characters are both FAR more superior than AI. Same is said for voice acting/talent, artwork, etc etc etc. AI just isn't as good as real talent by people who know what they're doing and execute things really well.

[limber nexus]

exactly

[velvet crest]

I ONLY use AI as a hobby and nothing more.

Oh, music as well. AI music is nowhere near as good as real music.

[limber nexus]

me too

that too

[velvet crest]

I don't post AI content to YouTube or ANYWHERE publicly anymore to respect those who are against AI.

[gentle trench]

honestly wouldn't suprise me

[lost cape]

There are to many ai generated pictures on my system!

If only I could share a tool that fixes that problem on #1359898289335566570

Woe is me!

[frail tusk]

Is applio rvc really better?

Ive never tried SVC, but I’ve found rvc has a slight muffle to it because it seems to smoothen out sound

I used a 30 minute super high quality dataset recorded on a Se2300 microphone, but it still doesn’t sound crisp and clear

[gentle trench]

possibly could be using a old version of applio unless u use the online version

or could be u using a bad pretrain

like titan

[frail tusk]

No I’ve used the online version

[gentle trench]

ah

[frail tusk]

Have you got an example of your own audio with singing from a clone

I’m trying to test something to see if it’s just me

[gentle trench]

I have many models I've made

[frail tusk]

Do you have one clip of it singing

So I can compare to mine?

[gentle trench]

u can look at this post here I made, I don't usually make models of regular human voices tho
https://discord.com/channels/1159260121998827560/1453130384760897676

[zenith plover]

nah

[scarlet flume]

guys I want to make a video through AI , just by texting but I don't want to pay lol

what do you guys recommend?

[marble fractal]

sora ai lol

[misty frigate]

titans not a bad pretrain

[gentle trench]

it creates harmonic distortions and kinda makes models sound bigger

[misty frigate]

Well, based on my experience, it trains fast and is one of the more stable pretrains I've tried.

[misty frigate]

Also, the distortion you've been facing could be due to the config of contentvec

[odd knot]

it struggles with higher notes

imo

[eternal geyser]

hello

[compact nebula]

Imao

[gentle trench]

Try legacy core 1.5

Train the same model and see how different they'll be

I personally use the 32k hifigan contentvec version

[wispy palm]

anybody having issues with collab applio, saying no interface running?

[nocturne temple]

don't use it

[gentle trench]

True kaggle is better

[nocturne temple]

local machinery

he's listening to rein raus

https://open.spotify.com/track/4fPrxIohi60EZfNy2TlD44

[covert lake]

Elaborate in #1192011222023950368

@maiden terrace don't promote

[gentle trench]

Ratchet from ratchet and clank

[nocturne temple]

yes

[gentle trench]

I think

[gentle trench]

Nice

[nocturne temple]

I am posting this stuff on civitai.com

[proper flax]

Hey everyone, I help busy people with AI generations I have my own software, contact me for more info in DM

[worldly oasis]

theres no natalie nunn voice model??

[gentle trench]

If it's not in https://discord.com/channels/1159260121998827560/1175430844685484042 you could ask someone who takes free commissions or just make it yourself 🤷‍♀️

[maiden terrace]

Yeah my bad, meant to send that to myself so I can book it 🤣

[proud marsh]

since weights is going to shutdown, what whoul be the best way to preserve/save the voice models on it?

[gentle trench]

download them and move them to huggingface

[forest crypt]

https://tenor.com/view/cat-yawn-bite-interrupted-bored-gif-10217241

[native tendon]

why does it sound different when I import a voice like a booster?

[pale helm]

мне нужен эфект рации парни

помогите найти

[solar torrent]

This server is English only.

[odd knot]

this is an english-only server. we don't speak cyrillic here.

[pale helm]

ok

[solar torrent]

[odd knot]

oh wow we jinxed

[native tendon]

брат

хелпани

[pale helm]

can you help me find radio mod

[native tendon]

я кароче беру голос ивана золо и симпла а он звучит по другому

[pale helm]

brat ya tut sam niheya ne ponimayu

[native tendon]

ваще

я ели как сделал

это

[pale helm]

брат я брал ток голос доры а хочу сделать эффект рации на голос

[native tendon]

дорв норм работает?

дора

[pale helm]

да

[solar torrent]

If you two still speaking Russian in here, it's better to go to your direct message instead, or stay here and speak English; otherwise I'd just call a moderator to think about your actions.

[pale helm]

ваще заебись

[pale helm]

ok sorry gues

can you help me find ai effect?

[solar torrent]

For help about the "voice changer" as you implied, better go to #✨│ai-help.

[native tendon]

ls

[dense phoenix]

Hey everyone 👋
I’m Bacem, founder of RealiWrite.com

I built it because my own AI drafts kept getting flagged or sounding robotic.
Happy to answer questions about AI detection, humanizing text.

[twilit peak]

Been off for a year, is Applio still the best for free cloud training?

[solar torrent]

Applio RVC is both a local program and an online notebook (Colab/Kaggle). Yes, Applio RVC does exist as the online one.

[twilit peak]

Yeah did something better come out in the past year?

In terms of computational speed/limit or model quality

[solar torrent]

Applio RVC can now use third-party custom pretrain models for training a voice model.

[balmy marsh]

Hai can someone please tell me their favorite websites to get speaking models?

[heavy brook]

Hey guys
what's the best pretrain to use if you want to make a singing clone

GPT is recommending either singerpretrain or Titan, but someone said Titan is bad

[elder willow]

somem dude said

ohh AI is failing!! noone likes it!!! openai will go bankrupt!! and i said

if it was actually full on failing and noone liked it

then the Millitary industrial complex wouldn't be INSANELY interested in it

[gentle trench]

Use legacy core 1.5

https://discord.com/channels/1159260121998827560/1453062294446805244

There are many different versions of it, I use the 32k hifigan contentvec version

[heavy brook]

Is legacy core 1.5 better than the default one applio provides? alsowhat about v2.5? is that any better?

[heavy brook]

also why 32k hifigan over 32k refinegan?

[gentle trench]

Refingan is quite new so I'm not sure if I want to use it yet over smth I'm used to

[gentle trench]

Legacy core 1.5 is an improved version of 1.0 even tho the number is lower than 2.5 it's better from my testing

The og pretrain is just a little worse than 1.5 tbh

But always good to test on og pretrain if your model comes out bad on legacy 1.5

[desert remnant]

<t:1769101593:R> does anyone want to chat?

[gentle trench]

I'd rather talk to Nvidia tbh

[heavy brook]

can i use it in applio?

I can't seem to see where to put in the files

[gentle trench]

Yes you put the two G and D files in applio

Go to the download section and scroll down to pretrain, click customize and copy and paste the D and G links into their respective slot

[heavy brook]

ah ok i found it i think

there are some pretraining models already there

singerpretrani

is that maybe any better than the legacy v1.5?

[gentle trench]

No those are old

[heavy brook]

thanks

[gentle trench]

Only current pretrains like legacy 1.5 or klm would be recommended

[minor topaz]

I did everything right and after I click on the start button in the CMD it shows warming up... generating sola buffer... and it does not start can I get some help?

[heavy brook]

google drive paths?

[gentle trench]

?

No it would be in applio

[heavy brook]

can i dm u

so i can send you a screenshot

[gentle trench]

Sure

[visual rampart]

Anyone want to vc

[limber nexus]

<@&1159293140440723499> #🌇│ai-videos message, #💻｜programming message, #🤖│bots message

[spark falcon]

Who is looking for a skilled AI engineer?

[dark rover]

I have seen those posts before, and they all look the same. It's likely AI generated, which makes it spam.

I am pretty sure that if you take a screenshot of these posts (before they are removed or whatnot) and then see one again here later, you'll notice how similar it looks.

They're all engineers, all full stack nonsense, all end to end ai solution providers, etc. The method of writing and word choice basically isn't human; it's advertisement language.

Either its a show off post (self promotion) or they're trying to manipulate someone else into replying to them (likely to extract money).

"I worked with Y" is kind of useless information for most people.
"By the way, did you know you can do X with Y if you set up Y like Z?" -- that is useful and also immediately proves you have some knowledge instead of it just being a claim.

[frank bough]

Did anyone ever tested diff-vc ?

[odd knot]

diff-svc is VERY old.

you need a LOT of data for the model to sound good

[frank bough]

Not svc

VC

[odd knot]

still

diff-vc needs a lot of data for the model to sound good

[frank bough]

What about zero shot version ?

[odd knot]

idk i haven't tested diff-vc

[dark rover]

<@&1159293140440723499> concerning people who post messages like:
#✦│chat message
#🧬│ai-chat message
#🧬│ai-chat message
#🧬│ai-chat message
I think you should consider these spam. They're diffidently AI generated posts (botting), and practically useless.
Don't exclude that last one: it is simply an editted version with a new prompt, causing a typo. Notice how it should be 'and practical ai integration solutions designed for"

[gentle trench]

There's a lot of people in here I swear aren't even people they must be bots

[dark rover]

I might need to make a suggestion thread about it I guess; that is if there is nothing staff can do about it currently

[gentle trench]

I mean just ban anyone who posts those dumb scam images that everyone of those bots posts

Wtf was that

Spam jumpscare

[elder willow]

-colab

[pine acorn]

📒 Google Colab Notebooks

Google Colab is a Cloud (Remote Good PC) Service. While the Free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

• **Applio**

by IA Hispano
Google Colab

• **RVC Mainline**

by Hina
Google Colab

• **UVR5 NO UI**

by Eddy
Google Colab

• **UVR5 UI**

by Eddy
Google Colab

• **Wokada Deiteris Fork**

by Deiteris & Hina
Google Colab

• **Hina's Modified Original Wokada**

Google Colab

• **RVC-AI-Cover-Maker-WebUI**

by Shiro & Eddy
Google Colab

• **FaceFusion UI**

by Nick088
Google Colab

• **FaceFusion NO UI**

by Nick088
Google Colab

• **Music Source Separation Training (Inference)**

by Jarredou & Makidanye
Google Colab

[gentle trench]

Kaggle is better tbh

[scarlet axle]

I just joined, and tried to download a Pearl model (from sponbebob) and it said that Weight is being shut down in March of this year. Just wanted to let ya'll know. :(

[worthy coyote]

do you have a harmony isolator?

i'm looking for it for months i still haven't found it

[cold hatch]

HEL THE FUCK LO?? WYM WEIGHTS IS SHUTTING DOWM

[solar torrent]

I already know about Weights, of course so many times, but there's #1175430844685484042 for voice models.

[scarlet axle]

sry

I just wanted to help by letting ppl know.

In case they didn't know either

[gentle trench]

yea it's shutting down in march, the site removed the ability to upload models so it made since they were shutting down

replay will still work tho

[cold hatch]

Thanks twin i js asked someone for the installer

If they dont respond
Do you have it?

[gentle trench]

I sadly don't but def ask in the official discord for weights

[cold hatch]

Tyy

[solar torrent]

I use Applio RVC and UVR5 in separate environments. The question is, how do I tell them?

https://tenor.com/view/cat-monday-left-me-broken-cat-talking-cat-gif-4255367456898854409

[gentle trench]

You tell them in the most complicated way possible

[finite orchid]

Hello everyone

[toxic gazelle]

Haven't been here 1 year. Has anything improved regarding ai covers during this time?

[left sphinx]

<@&1159293140440723499> crypto scam

[bold lily]

instead of scam bots can we get hot men repost bots

thanks

[solar torrent]

Send them scammers with Skibidi Toilets.

[edgy light]

Info about Legacy Core 2.9 pretrain? There's no info for that.

[odd knot]

legacy core 2.9 is a speech pretrain trained on ears data

it's currently available in 48k tho it has some raspy problems the creator couldn't fix

[edgy light]

And embedder for Legacy Core 2.9?

[odd knot]

contentvec hifigan

i wouldn't really recommend using it if you don't have pure 48k data

[edgy light]

oh, okay waifu girl

[odd knot]

https://tenor.com/view/ado-cupcake-ado-cupcake-ado-plushie-plushie-cupcake-gif-12188714983951951436

[turbid zinc]

hi How can I download the voice model if registration is disabled and the page doesn't load?

[gentle trench]

???

[minor blade]

Ig he's referring to weights

[gentle trench]

maybe

[river gust]

(ok i'm pretty sure there won't be a single soul in here who cares but)

(at least the devs in here can relate to that feeling when something grand finally works)

my modem can now send T.4 ECM using V.27ter yayyyy

......thanks to ChatGPT, of course, but.......

still, it works, yayyyy

[gentle trench]

idk what that even means lol

[river gust]

i'd love to show you, but

sadly, (a) i can't stream here, (b) it might be boring for non-nerds

but of course, there's a caveat to (almost) everything.... i can only send payloads of up to 65535 bytes with ECM enabled

@gentle trench remember those old internet modems from way back when? yknow, those things that made the meme sound

i made chatgpt code something similar... instead of an internet modem, it's a fax modem

[gentle trench]

Ohhhh

That's cool

[river gust]

i can literally send faxes with my PC now

spitting straight fax™

the technology is very much obsolete of course, thanks to email etc., but

the cool part is being able to reproduce it, here, in Visual Studio, and seeing it work in action
PS: and in part, the journey: getting over obstacles on the way, adding features, refining them, seeing it all come together

(btw, at least in my overall line of work, fax is still very much used regularly ..... it has its benefits)

i'd like to conclude by saying:
in comparison to today's internet, audio-based modems are EXTREMELY slow (who would have thought).... but let me tell ya, i've gone through parts of the documentation, and it's clear some real geniuses were behind those machines

it's pure math madness in there...... those things had some serious thought put into them

i can't imagine how sophisticated today's routers etc. must really be under the hood, seeing how things were back then already

[glossy nest]

Yo

[river gust]

The result
https://simple-fax.de/Downloads/testfax/6973f6a5d207a104486.jpeg

[gentle trench]

https://tenor.com/view/ena-ena-joel-g-ena-dream-bqq-dream-bbq-coral-gif-10464623215235539382

[river gust]

https://imgur.com/gkmlUws

what am i even looking at

[gentle trench]

coral glasses from Ena Dream BBQ

[river gust]

ok i didn't see much but

that game straight up looks like a fever dream

[marble fractal]

awsome

[feral solar]

whats the best girl model

[bold lily]

self promo

you're going to be banned in 5 minutes.

[gentle trench]

https://tenor.com/view/car-garn47-47-cat-game-gif-18225718487563422536

[limber nexus]

Great suggestion

[gentle trench]

truly

[solar torrent]

This thing. https://static.wikia.nocookie.net/oniichan-wa-oshimai/images/3/3b/MahiroAnimeCasual.png/?format=original

[ionic leaf]

anyone knows how they are remixing songs into completely different genres? I saw YE songs getting reminxed into punk rock and it goes hard as hell

[solar torrent]

There's FL Studio.

[solar torrent]

https://tenor.com/view/silly-gif-1765839720796790298

[formal carbon]

anyone have ai girl voice in french?

[low sinew]

😄

S0N OF A B1TCH!

[solar torrent]

Kasane Teto likes baguettes.

[formal carbon]

[steady lagoon]

@worldly wigeon you is a fucking bitch just ban me already from yo shit ye server

I literally ask you to ban me and you mute me for 500 days and call me a ret@rd

[worldly wigeon]

Disable the guild restore from the notification you get in your DMs and leave the server

It’s not a big deal

[steady lagoon]

mf just ban me

[worldly wigeon]

No

[steady lagoon]

😭

you desperate for members or smth

[worldly wigeon]

Nope

I’m just not giving you the satisfaction of being banned

So you can do it yourself for spam pinging me at 8 AM

[edgy light]

Stop it!

[solar torrent]

@steady lagoon @worldly wigeon You two might better have take this conversation to y'all direct message instead of here.

[steady lagoon]

lmao I'm done already I just wanted to say that one thing lol

i don't even know him

[worldly wigeon]

You’re like 40 minutes late the conversation is already over

[steady lagoon]

true

[river gust]

i did not raise either of you to act this way

[ebon spruce]

is there any other good ai video, without credits and more except mindvideos. because they used to do 4 videos a day now its 2 a day on a account. 😢

[cloud kettle]

Welp weights.gg is getting shutted down

So uh, is there any alternatives that are like weights?

[gleaming oak]

This #✦│chat message

[gleaming oak]

Video X https://huggingface.co/spaces/liuhahi/wan2-2-fp8da-aoti-preview
Video with audio
https://huggingface.co/spaces/alexnasa/ltx-2-TURBO
To extend your usage time x8 without paying, check out the last part of this guide
https://huggingface.co/datasets/TestOrganizationPleaseIgnore/help/blob/main/gpu_help.md#what-is-zerogpu

[ebon spruce]

none are working for me, any others, some need of thoses need images

[restive bluff]

whats the best egirl model

[gentle trench]

Just for you Mr catfisher

[edgy light]

I said no e-girls allowed

[lean jolt]

Hey anyone got an idea how to edit in audacity so that a repeated word within a word by a clone is smoothed out

So for example I have my clone saying flame - but because of a backing voice, the clone says ‘flalame’

I’ve tried just deleting the extra ‘la’ but it causes a popping sound as the audio suddenly drops and picks back up

Also having same slight issue with trying to do a fade in / out

But maybe I’m not doing it right

Anyone have a suggestion on what the best way to do this is

[strong niche]

yooooooooooooooooooooooooo

SUPPPP

[wind totem]

How do people make these videos!!

https://x.com/slatdontmiss/status/2015115265416077347?s=46

[gleaming oak]

ltx2, wan2.5, grok, sora2

[strong niche]

dis server dead

as heck

[marsh ermine]

Hello guys

What is currently the best AI for upscaling images?

[desert perch]

hello! can someone help me do an AI track. i want to change the singer

i have the model

but idk how to do the rest

[solar torrent]

According to your name, Kasane Teto.

[gentle trench]

true

[solar torrent]

I was sleeping.

[jade hull]

Hello

[limber nexus]

[river gust]

best free one, until you find something better
https://unlimited.waifu2x.net/

[stable fern]

[odd knot]

chainner + illustrationjanai for anime images

trust me

[rough bridge]

Yo chat

Gimme some free text to video and text to image tools

I got none good

Everyone asking for money

And sora 2 is bs

Veo is paid too

Any good ones?

[woven python]

Hello, can someone provide me with a female Turkish voice model?

[compact spire]

Tensor and PixAI

[worldly oasis]

guys i forgot everything how do i get to that gradio thin i forgot

thing

so i can make an ai cover

like is there a link

[forest basin]

sup

[solar torrent]

No.

[sterile star]

hi

[bronze pasture]

sup

suuuuuuuuuuup

[tame void]

Hi guys

How is everyone

[drifting ivy]

hi there

can anyone please help me with the best girl voice model?

[gentle trench]

here ya go mr catfisher

[drifting ivy]

come on buddy

[gentle trench]

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

[sinful cloak]

RendrFlow

[young yacht]

What is the best site to use to create voice models now that weights is going away?

[gentle trench]

applio

it's always been better

[young yacht]

Is it a program or a website?

[gentle trench]

it can be used both locally on your computer or on a browser

-rvc

that that command in https://discord.com/channels/1159260121998827560/1159290139609137264 to get more info abt it

[young yacht]

Thanks! I'll check it out when I get home!

[gentle trench]

no problem! it's fairly easy to learn

u can make covers and train models with it

[young yacht]

I've been using mangio rvc to make voice covers. Could never get the train models function to work though.

[odd knot]

mangio rvc is REALLY old

and outdated

[sharp current]

so im new in all these real time voice changers. i tried some and the results vary.
i have GTX 1650 with 4gb vram so i know im severly underspeced.
thats why i saw beatrice v2 and its was fast but there's no voice model for that. im still very confused as to what actually is what. i currently have beatrice v2 from W-okada github and vonovox. i want to go beatrice v2 since its fast but there's no model for it and its doesnt work with other model trained with RVC

[gentle trench]

beatrice is supppper outdated

what u have is old and most likely from an outdated youtube tutorial

[sharp current]

then what are my choices for now?

[gentle trench]

for you, either wokada tg fork or vonovox

don't look them up onlline btw you wont find the right stuff

[sharp current]

i have vonovox and it works but its still heavy on my system

[gentle trench]

o

[sharp current]

and wokada tg fork open browser and that eats all all the resources

[gentle trench]

hm

[sharp current]

seems like vonovox is the onl way

[gentle trench]

I guess you're stuck with vono yea

[sharp current]

anyways, thanks for chiming in. also now where can i get voice models since weights is gone

or can i still download its model from somewhere else

[gentle trench]

right here in this server https://discord.com/channels/1159260121998827560/1175430844685484042

this is the only place with high quality models tbh

[young yacht]

Does it normally take a long time to train a voice model on applio?

[umbral sun]

can anyone help on #✨│ai-help

[uneven scroll]

Ai

Anyone got a good dommy mommy voice changer

Im a girl

I just need the pitch

[gentle trench]

just follow a guide lol https://www.youtube.com/watch?v=aLgwc9acGOI

[bleak slate]

@wet current Check Dms I remember you from Roblox lol

[slender grove]

Are they gonna make a different ai weights because it’s getting removed

[gentle trench]

nah

[solar torrent]

Only if they could make one a replacement. It's now a cursed curse for "freeware" AI website services, as much as free users popping around, look at Weights website of early days as example.

[solar torrent]

Usually over hours, but also depends on your GPU specs.

[solar torrent]

https://en.touhouwiki.net/images/5/5e/Th15Doremy.png

This character is named Doremy Sweet.

[uneven scroll]

Ohhhh nicere

[sinful cloak]

Hey everyone, Anyone want to help me stress test a local AI image Upscaler? RendrFlow handles batch processing and 16x upscaling entirely on-device to keep your photos 100% private. DM if you want to give it a spin!

[solar torrent]

Try make it 16384x16384.

[sinful cloak]

Yeah you can make if you have power processor on Android

[rugged island]

Do you have a git repo or is this already a compiled app to install and run?

[marsh flicker]

Hey everyone

does someone have product hunt?

[small hazel]

why does my mic sound better

on voice mod then ai thing

[chilly plume]

Where can I find the DMR V2 pretrain model?

I can't download it on the #1235952130855010365 channel

[slow ice]

Hi guys is there any AI model that provides a free, unlimited use API?

[gentle trench]

If you're training a model I'd use legacy core 1.5 over anything else

[sinful cloak]

It is already in play store search Rendrflow

[fresh gust]

Is there another bot like astralabs?

[formal obsidian]

Is everyone here AI?

[gentle trench]

Yep beep boop beep

[teal estuary]

anyone here knows about ai automation for whatsapp chatbot

[tawdry ruin]

i just need tokens by refering people

[still sigil]

@tawdry ruin no ads

[tawdry ruin]

oh mb

sorry

[wind totem]

Beep boop

[gentle trench]

download from huggingface

[lusty steeple]

anyone open to collab for ai music? im a songwriter and love transform my lyrics into ai hits

[gentle trench]

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

idk what that even means

click that

in the model section, any model

[gentle trench]

Models made on weights aren't good, but if they were just uploaded on weights but not made on there then idk

Kinda screwed

[elder willow]

hi guys! is there a new update for W-Okada Voice Changer? or still same?

[weary yoke]

Is wavbots rvc model up to date or is there a newer one?

[sharp current]

which one is lighter?

[weak valley]

Had to reinstall windows

It's taking longer than expected

[solar torrent]

Asus ROG.

[solar torrent]

I'm not sure about "WavBot" but I know about RVC (retrieval-based voice conversion).

[solar torrent]

There are W-Okada voice changer** forks** that are implemented from the original mainline W-Okada. There's also Vonovox the alternative voice changer as well.

[solar torrent]

I can mix tracks, make simple drums patterns in DAW software, but I'm not sure about lyrics.

[elder willow]

ill check out vonovox! thanks for replying!

[covert lake]

Wdym with wavbots?

You can search rvc ai voice models at:

• https://discord.com/channels/1159260121998827560/1175430844685484042 (main source)
• https://www.weights.com/it/models (login required, shutting down on March 31st 2026)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• https://voice-models.com (just scrapes the huggingface link from the voice-models channel)
• 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
• Ask a free request in https://discord.com/channels/1159260121998827560/1159289738314919936
• Be aware that we don't allow any paid comms in the server

https://tenor.com/view/l-lawliet-death-note-gif-21936347

Not sure but you could try logging in and download the ones on weights till March 31st 2026 iirc, otherwise there's nothing else you can do

[deft iron]

Looking to buy a new GPU and was considering the 9070xt

Does anyone know how good the ai training performance of it is compared to nvidia?

Or is AMD still far behind in compatibily and performance

[frail tusk]

Sure

[vivid breach]

Where’s a support channel

[gentle trench]

for what?

[vivid breach]

Voice changers I guess

[gentle trench]

you'd just need to ask herehttps://discord.com/channels/1159260121998827560/1159290139609137264

the voice changer's name is w-okada btw

[thorn stirrup]

Hi guys

Anyone heard of Silas Spectrum?

[dark rover]

-googles- ... -finds nothing related to AI-

[wheat seal]

is there anyway to like submit my ai voice models on here? I know how to make them. but where do I send the files to get the model maker role?

[gentle trench]

right here https://discord.com/channels/1159260121998827560/1452279591317278840

[wheat seal]

I don't have access to that for some odd reason or is there smth I'm forgetting?

found it nvm

[thin mist]

@covert lake Hey can you download 2 models for me from weights? I don't have a login either

[coral marsh]

https://tenor.com/view/not-everyone-has-an-iphone-gif-7021079305409267183

[solar torrent]

Check if there voice models you might be looking for in #1175430844685484042.

[crimson token]

bro

grok is goated

https://x.com/i/grok/share/077c8f33d5074632ad6f5f67d03243dc

[gentle trench]

I hope it's not what I think it is

[crimson token]

its not

its not cp

call of duty, points

[solar torrent]

I didn't ask about your opinion.

[haughty cave]

Damn I talk to chatgpt and grok than I do to actual humans

[solar torrent]

I talk to real people more than you typing to a chatbot.

[gentle trench]

real people are more fun than robots

[lusty steeple]

perfect may i dm?

[solar torrent]

Only "free" request.

[prime aspen]

best voice module for a deep voice for trolling

[solar torrent]

The voice model for trolling is "Killer Fish from San Diego".

[prime aspen]

nothing came up?

[solar torrent]

Because it doesn't exist yet in #1175430844685484042.

[prime aspen]

do you know any modules with deep voices idk what to search up

[solar torrent]

[peak quiver]

https://discord.com/channels/1159260121998827560/1163571683848900629

it doesn't work now ?
Facefusion on lighting.ai with GPU on but it says
facefusion.py run: error: argument --execution-providers: invalid choice: 'cuda' (choose from 'cpu')

pls

[covert lake]

rip then

[peak quiver]

Ohh btw , Is there still working facefusion on cloud pls

[gentle trench]

I fear what you need face fusing for..

[unkempt saddle]

@gentle trench Hi, It seems you're an expert here. I'd like to ask if there's a way to add languages ​​to a model that only supports English?

[gentle trench]

Models trained on English can speak in any language, same way around

They're basically filters

It's how they can sing and stuff exactly like the singer and character at the same time

@unkempt saddle

They might not sound perfect in a different language compared to the one they were trained on but that's due to rvc limitations ect

[bitter helm]

It's seem, when I take a model trained on english, when I speak french the voice will have an english accent

Do you have any recommendation, I'm looking for french trained voice

[unkempt saddle]

I usually use RVC v2 realtime. For example, I often choose a model that supports Japanese and set the index rate to 1.0 (because I want the model's voice to sound as close to the model's as possible). When I speak English using this model, the sound comes out more like Japanese words than English. So I'm wondering if there's a way to train that model to sound more like English.

[gentle trench]

I don't personally know as I train models on English only pretty much, maybe the spy from TF2 🤷‍♀️

[gentle trench]

That's due to the index, don't turn it up at all it's better without that

Causes all types of weird stuff

Plus uses more GPU

[compact spire]

Hi

[bitter helm]

I got a 5090

Do you know any tutorial

So I can start training french model ?

I think it's the only way

[gentle trench]

Any YouTube tutorials are outdated, but use applio

If you go to https://discord.com/channels/1159260121998827560/1159290139609137264 and type "-rvc" you will get the guide for applio, applio can be used both locally or on browser which is what I use it one for my convenience

You'd also need to get audio of a French character or person off of YouTube or anything really and clean it using mvsep or uvr

[bitter helm]

Ok thanks

And what client do you advise me to use ?

Cause I'm still using this one https://github.com/w-okada/voice-changer/blob/master/docs_i18n/README_en.md

Do you know if there is a better ?

[gentle trench]

This is very old, I would switch to Vonovox

Type in -rt in that same channel where you got the applio guide and download Vonovox

[unkempt saddle]

Thank you, perhaps this is the limit of RVC, because I want the index rate to be 1 since if it's set to zero, the accent will sound more like... me.

[bitter helm]

ty I'm doing it

[solar torrent]

RVC and W-Okada voice changer are two different programs. This specific W-Okada version wouldn't work with NVIDIA GeForce RTX 50 series because it's outdated and wasn't compiled for that GPU. For better working version, consider for Vonovox or Tg Develop's W-Okada fork instead. And for more help about the voice changer better go to #✨│ai-help.

[solar torrent]

The "index" the RVC voice model's accent file, because the task is CPU-intensive, the performance drop is often unnoticeable in non-realtime one like Applio RVC, whereas realtime voice changers like W-Okada and Vonovox might struggle if "index rate" is set to a non-zero value.

[bitter helm]

@gentle trench Hello

Can I add you in friend ?

[edgy flare]

Looking for a Volunteer Project Manager for AI Startup 🚀

Hi everyone!
We’re building an early-stage AI startup and are looking for a volunteer project manager to help coordinate our team and drive our MVP development.

If you:

Have experience managing tech or AI projects

Enjoy fast-paced, innovative environments

Want to gain startup experience and shape something from the ground up

…then we’d love to hear from you! DM me if you’re interested and want to learn more.

[drowsy inlet]

ai will give me job 🥰

[gentle trench]

Sorry I fell asleep last night, why?

[regal bobcat]

what the best ai vm to sound like young girl

[gentle trench]

I don't like where this is going

[regal bobcat]

🙂

[median marlin]

is there a channel where i can download the softwares or whatever? i remember it being there a couple months ago

[molten gale]

❤️💐

[gentle trench]

Which softwares are you looking for?

[median marlin]

the ai voice changer, its been a couple months so i completely forgot the name of it

[gentle trench]

Ohh, w-Okada

It's dependent on what gpu u have but I could get you the right one

[median marlin]

5070

thats alright i think

[gentle trench]

Alr bet, go to ai help type "-rt" and go to the guide that says Vonovox, that's the newest best real-time ai voice changer for Nvidia

[median marlin]

appreciate it, thanks 🙏

[gentle trench]

Np! If u need help u can text me in that channel as well

[bitter helm]

Can you help me to train a model

With Applio ?

I got a nvidia 5090

[river gust]

i actually have a similar question now

never thought i'd be asking this, but

if i wanted to create an AI that plays a game based on what's on-screen, what would be an effective way of doing that?

i was thinking about creating an AI player for Panel de Pon, or its numerous namesakes, in this case Pokémon Puzzle League

fairly simple game, so i would think it'd be doable.... but

i've tried before, and it totally failed

even just screen analysis was a problem

[gentle trench]

There's a guide kind of if you type -rvc in ai help but I currently am not home so I can't help really

Remember to use applio

[bitter helm]

Yes

i'm currently using applio

But I struggle

[sinful lily]

how did you guys get into AI?

[vivid spade]

Hi everyone, I'm using the RVC WebUI that you can access from computer files, it's been a while since I used it, but I just tried to import a voice model and it's not appearing in my inferencing voice dropdown even though I put the INDEX and PTH files where they should be? is the UI broken?

[river gust]

cough

nobody who can give me pointers or anything?

[dark rover]

try asking in #1192011222023950368 or #✨│ai-help
this is a more general chat

[vivid spade]

okay ty!

[dark rover]

Tools exist that do not rely on AI that can do this.
Botting is as old as runescape, if not older.
using AI to autoplay is probably inefficient, not just on the power consumption level.

Take a glance at MMORPG servers for example, the server runs a separate NPC server, which loads in fake players (npcs), that has their own inventory and such. (usually it also handles mobs). More complex variants have such fake player actually join the player party and follow the player into instances.

Screen control is also something that has long been invented (also from a time before AI).

[river gust]

well sure but

being an amateur, where would i get started

[dark rover]

I'm not going to help you with that, because of principle. I don't want people to have an unfair advantage over others in competitive play, or ruin the economy inside an online game through automated systems.

[river gust]

lmao the game i'm talking about is entirely offline rumi

and i'm very good at it myself

i just want to, like i always do, marvel at the prowess of AI, or even just simple bots

i've always wanted to make a bot that's better than the ingame one

or am i in the wrong channel for this

you know what, i probably am, not least because of a lack of response

later.

i'm doing something with chatgpt, and allegedly it has been "thinking" for more than three hours at this point

[ebon spruce]

My new Batwoman video is out plus I used AI video and images

[dark rover]

You know what it is, any tool that can automate a singleplayer game by reading the screen can also automate a multiplayer one. Everytime you sleep you can let your system grind an mmo or whatever- gathering points, which would give the user an advantage over those who don't have the same means. Such means also becomes worth money, because people like cheating and pretending they are playing on the most difficult settings at the same time, which- yeah- I don't see offline botting tools as good either for that reason.

I think its fine for a game designer since they are creating a game, but when players just want to be lazy or game when others cannot, (during sleep, work, etc.) or control 20+ systems at the same time to the advantage of one person, then no

[versed lotus]

how can we do live video for long time like this: https://x.com/pounddz/status/2016478185571942853/video/1

if i want to go on omegle and be like ja ja binx or something

[lavish mauve]

Hi I'm looking for the biggest Ai factory being constructed or planed by like 2030, and I found this is the largest on a single site:
Project Jade:
located in Cheyenne, Wyoming.
to scale up to 10 GW by 2030
by Crusoe Energy Systems in partnership with Tallgrass Integrated Logistics Solutions LLC

I just need some fact checks if this will be the most powerful Ai or something.
or if it is better to have like a distribution of multiple sites like stargate will be..

[lavish mauve]

also is anyone using Clawdbot (Now Moltbot)?

[ivory karma]

can someone help me fix my virtual cable

its not working

[untold wave]

yo

new here

[untold wave]

Discord notifications either save your life or ruin your sleep.

[limber nexus]

Real

[untold wave]

thats it

[untold wave]

how are you doing out here?

[limber nexus]

im pretty damn good, wbu?

[untold wave]

thats good
but new here

[limber nexus]

Welcome

[untold wave]

nie meeting you

[limber nexus]

likewise

[untold wave]

will you mind if i send you friend req

[limber nexus]

Not at all, you don't seem like a scammer or bot

lol

[nimble hare]

hey, does anyone have any experience training sdxl lora's?

[limber nexus]

i know a fwe guys in #🏙│ai-images has trained SDXL LoRAs

[nimble hare]

perfect ty!

[limber nexus]

you're very welcome

[nimble hare]

nice to see this server is still running i havent been here in yearss lol

[limber nexus]

lol, we revived it at one point

then we killed it again

[nimble hare]

yeahh it definitely doesnt seem as active as it used to be

[limber nexus]

Real

[dark rover]

might want to ask that in#✨│ai-help , but for now just reinstall vaclite and restart your computer, then reassign it.

[pastel meteor]

何

これをテストしてみるよ

[brittle tinsel]

Hi, I'm new here. I was just trying to download a model, but it looks like Weights.gg is shutting down. Is there any other way to get the model? Does it work the same if I use the files from the Huggingface link?

[dark rover]

If the author is the same, chances are the model is the same as the one stored on weights
You can ask the author if they are here directly about it

[gentle trench]

yea they work the same

[royal fulcrum]

test

[buoyant dome]

Yoo

[weary radish]

bruh I cant see the ai models

[royal fulcrum]

how to use ai

[still sigil]

#1175430844685484042

[gentle hamlet]

🥸