🧬│ai-chat

[drowsy hawk]

4080S
also have a 32 gb of ram and nvme drive

[solar torrent]

If you have a GPU that's older and slower than GTX 20xx series in your PC, I'm not sure what to do with AI other than RVC and W-Okada.

[drowsy hawk]

check the message above again my lil bro:)

[solar torrent]

Well, that's a pretty decent specs your PC has. No, I was just too slow to respond.

https://cdn.discordapp.com/emojis/872078487639453707.webp?size=48

[polar flax]

on the other side, you could still train flux loras, though with more limited settings than for 20+ GB cards (the irony that 5080 will still be 16 GB)

[drowsy hawk]

at this moment, tweaking some of the settings, I can freely use flux to generate whatever (the largest one). Takes about 30 secs, but still pretty good.
Anyway, planning to buy 5090 later

so what's about the main question?

[covert lake]

That's good, ofc not as much as an AI-GPU

for LLM managers, my opinions are:

• General LLM:
  • llama.cpp: Basically what every other LLM manager is based on, can be used for servers and it's the one with most updates, but NOT user-friendly and harder to use
  • Ollama: Built around llama.cpp, very user friendly and easy
• Roleplay:
  • Koboldcpp: llama.cpp fork, easy to use, not the best UI tho
  • SillyTavern: best UI and functionalities but a bit harder to setup as you would need to use an API (such an online one or using Koboldcpp locally)

I'm not sure on which local LLM would be the best for coding
But you could try Llama 3.3 70b and maybe Deepseek coder v2

Btw I don't do much locally, I played aroung with koboldcpp & ollama with Llama 3.2 3b as I'm on Laptop

[drowsy hawk]

I am sure that 70b will not run on my pc. Or there appeared some tweaking to offload some pc parts when using ai sacrificing speed?
Also, which are best setting for tokens, penalties, (maybe system prompt?)

[covert lake]

Also Gemini 1.5 pro & 2.0 flash are good for coding imo with a shit ton big of context lenght

You can use them for free with limits on the aistudio.google.com but i'm guessing you want something local

[covert lake]

GGUF Quantization could help https://huggingface.co/bartowski/Llama-3.3-70B-Instruct-GGUF

It's a technique that 'compresses' the model weights, there are different versions of it, from smaller to bigger, it's explained for each on the model card

[drowsy hawk]

anything cloud is good too (since chatgpt is not working for me anymore, trying to find whatever best alternative are)

[drowsy hawk]

thanks:)

[covert lake]

Personally, I find myself good with gemini models, especially for their big context lenght
they give really good limits imo (1.5k requests daily for flash models, and 50 for pro models), ofc u can pay to get bigger limits

I heard some say claude is good but I never tried it

[drowsy hawk]

Sadly, claude does not gives ability to test heavy models for free

[covert lake]

groq also gives some LLM models for free https://console.groq.com/docs/rate-limits, you could use their console to use llama 3.3 70b for free, except on the free tier the maximum context lenght is 8k so wouldn't work for super super long code

it gives around 15k requests daily per models, reminder this is mostly made for APIs, but you can use their console too

[drowsy hawk]

hom much lines of code or characters is approx?

[covert lake]

ehh not sure about that, I think it depends by the tokenizer of the LLM

[covert lake]

Btw I wanna emphasize that all that LLMs do is just Text Prediction, they don't think nor reason about code

So ofc they aren't as good as a developer

[drowsy hawk]

No I mean, i was literally before full coding perfecty with chatgpt (literally 80% of the work), as I was already studying the “prompt science” for about 3 years, and I am a coder for 10 years

Now it just broke no matter what I do

[polar flax]

it's still useful for simple tasks, still it doesn't motivate me enough back to software engineering stuffs, where I have to deal with understanding client requirements

[drowsy hawk]

which from those two are better?

[covert lake]

2.0 flash is recently released and in preview/experimental mode

I would honestly go with 1.5 as it's more stable
But you can also just play around with both of those ngl

[chilly lake]

would be funny if intel drops 590 card with 32GB VRAM @ $500

vs nvidia's $5k 5090 (if you can even get it from scalpers)

[rustic epoch]

Hello! I'm looking for someone who makes models for free, to exchange something

Anyone who is interested should contact me in DM

[covert lake]

it will be 5k???

Ngl there's a pretty big difference stil from 4060/4070/4080 to 4090

[chilly lake]

"from scalpers"

[covert lake]

the 4090 is like 1.5k

[chilly lake]

you'll have 0 chance to snatch it before bots buy them all

right now you cant find 4090 for less than 4k

[solar torrent]

Why a new NVIDIA GPU will be $5000? That's a higher-end desktop PC price.

[covert lake]

why tf is the 4090 so expensive compared to a 4080

[chilly lake]

nvidia releases 5090 for $1999, bots snatch them and relist for 5k

[chilly lake]

bitcoin

[covert lake]

are u fr?

[edgy bloom]

Congratulations Nick088 [ITA/ENG] by Weights!

Your Charizard is now level 62!

New move!

Your Charizard can now learn Flare Blitz!

‎

‎

‎

‎

[covert lake]

for fucking crypto?

I swear u could find a 4080 super even at max 1k, then can't find a 4090 at lower than 1.2k

[solar torrent]

Is mining coin still a thing today? I thought it has just died out long ago.

[covert lake]

i seen the 4090 being sold even at 2k atp

[solar torrent]

eBay moment.

[covert lake]

@chilly lake btw which gpu would the best rn and that doesn't cost a kidney?

was thinking of getting a 4060 ti

[chilly lake]

there are a bunch of 4090 D chinese versions

24GB, but less shaders

similar to AMD RX 7900 GRE

[covert lake]

is that the less power one bc of a fight between usa and china for ai?

i remember hearing about it sometime ago

[chilly lake]

regular vs D

they are still around 3k lol

[chilly lake]

4060 TI(?) with 16GB probably, more memory helps with AI stuff

for gaming extra 8GB for extra $120(?) does not do much, but for running AI generations and stuff it is very beneficial

[covert lake]

yea would prob get that

[ancient swan]

used 3070 ti 16 gb

[chilly lake]

if you can find that

[covert lake]

I would mostly do school (coding), gaming & AI

and I need durability

[ancient swan]

yeah

[covert lake]

I don't know if to get used ones as I would need to not change it up in 2 years yk

[ancient swan]

if you want more power for less money but with good software support then used nvidia cards is the way to go

[covert lake]

mmm

[covert lake]

moral of the day: fuck crypto mining ong

i thought people stopped after they got scammed for the 10th time after buying an 'hawk tuah' coin and losing money

https://www.bbc.com/news/articles/c89xvjkzzyvo

[ancient swan]

doesn't that thing cost like 600 bucks

[covert lake]

basically everyone just:

gets famous
makes memecoin
makes its price rise
sells it
goodbye

[ancient swan]

in some places in europe

[chilly lake]

it was about 525-550 for 16GB version

[ancient swan]

bruh this is straight up legal scam

[covert lake]

i mean not as high as a rtx 4090

i was shocked checking the price

2k for a 4090 is kinda crazy

[ancient swan]

you can find it for 1600 bucks, it's a lot but, it's also unfathomably fast

[polar flax]

meanwhile Intel has got lucky for their B580 sold out like 9800X3D has been
https://www.techspot.com/news/106006-intel-arc-b580-graphics-card-successful-hard-find.html

[ancient swan]

titan level cards always cost a lot, and xx90 are basically rebranded titans

gtx titan cost like 1200 bucks at launch, which today would be like 1700 dollars probably

[ancient swan]

this is a great budget card

[polar flax]

it's like:
GTX Titan = 1090
RTX Titan = 2090

[ancient swan]

12 gb, 30-60% faster than regular 4060 both in raster and ray-tracing performance (if the game is already optimized)

[ancient swan]

yep, exactly

[polar flax]

hopefully Intel could also release B770 soon

[ancient swan]

they decided to rebrand it probably cus everyone viewed titans as semi-pro cards, when they were meant for gaming, so they just decided to fully market them as ultra beefed up gaming cards

but everyone for some reason thinks that xx90 are successors of xx80 or xx80 ti cards, when that's totally wrong lmao

and compare poor 1080 ti or 2080 ti to 3090 and 4090

[gray rover]

well.. ain't anything better in Poland tbf
13k that's like roughly 3.1k usd and 11.5k usd'd be 2.8k usd

all of that shit going is some serious fuckettry

that 7200 one is 4080

bruhhh

[ancient swan]

yeah, that's too much

[gray rover]

yea, doesn't help Poland was always so crap in gpu / nvidia hw pricing

basically shitting on people

[covert lake]

how is it for AI?

[ancient swan]

somehow even in russia it's not that bad, usually russian gpu/cpu prices pretty much match us prices

[gray rover]

for instance, I got my 3060 v2 for roughly 340 usd ( 1300/1400, in reality it shoulda been around 1k maybe 1.1k

shame

[gray rover]

Yea, I'd believe so

[polar flax]

have yet to see users running under ZLUDA, could be interesting to see comparison against 3060 and 4060

[ancient swan]

@covert lake what's your budget for gpu

[covert lake]

prob gonna get a pc with 2k (or a slightly bit more)

[ancient swan]

2k bucks?

[covert lake]

Euros yea

[ancient swan]

great

try to search for 3090 used

it should be like 650-700 euros probably

the performance basically matches 4070 super, but it has 24 gb of vram compared to 12 gb of 4070s

[covert lake]

thinking of:

• i7 14700F
• 850W charger
• rtx 4060 (8gb or 16)
• 16gb ram ddr5
• 1tb ssd

[gray rover]

why intel tho-

[ancient swan]

yeah

[gray rover]

ever since I moved from i5 10400 to r5 5600x ( ik, not all that great but still ) I ain't looking back at intel

hell nawh. life's been much easier ever since lel

[covert lake]

what's wrong with it ?

[polar flax]

AMD is dominating desktop cpu market

[gray rover]

Well, I guess it depends on what you go for

I am just biased so yea

But then, hearing all those recent situations with intels, I'd be careful

regardless of what's deemed as " fixed " or new cpu prod. line whatsoever

aftertaste remains imo

Just.. uhh, my opinion that is

[ancient swan]

ryzen 7 9700x (300 euros), used rtx 3090 (around 700 euros), 16 gb of vram is too little nowadays, get at least 32gb, it'll probably cost only like 150 bucks, the motherboard that fits the cpu for 150-200 bucks, and prolly at least some 1000w psu

[covert lake]

Mmm have never looked at AMD's CPUs ngl

[polar flax]

9800X3D is taking gaming king throne but it is also solid all-rounder for consumer use cases before 9950X3D arrives

[ancient swan]

intel's cpus currently suck not only because of problems with degradation over time but also because of their weird core and thread system that often goofs with applications

i often see people having to disable and enable some 0 core or some shit in task manager to make the application work as it should

so i would highly recommend getting some ryzen 9000 cpu instead of intel

• they draw way less power

maybe you won't even need 1000w power supply if you buy ryzen lmao

• some case that will fit the 3090 (it's a big gpu) 100-150 euros, 3090 used for 700 euros = 1700 euros

hell, you can even fit ryzen 7 9800x3d in the budget, it costs only 200 euros more than 9700x, the best gaming cpu in the world, that also has decent performance in other tasks like rendering, editing etc.

[willow grotto]

Guys how can I add 2 image models in 1 photo in weights.gg

[covert lake]

I would have to talk with my parents about this

[covert lake]

u just:

1. go to the image creation and click add lora (choose one and select)
2. click the + at the left of the previously added lora, to add the 2nd lora (choose the 2nd one and select it)
3. done

[ancient swan]

i would highly recommend getting at least something closer to the stuff that i've shown, cus the build that you listed is definitely too weak for 2k euros

and getting used cards isn't too bad

just be careful with the seller, make sure they don't put a brick instead of card in the box or something lmao

[covert lake]

I was afraid of that since I would want this build to last 5 years (without having to change the gpu in 2 years because its too used)

[polar flax]

remember when I got a 2 TB ssd for $90 in late 2023 (ADATA legend)

[ember mortar]

Where is the weights bot

[hearty surge]

yeah

[covert lake]

I can't really buy 1k gpu every 2-3 years

[covert lake]

@hidden grotto in #🤖│bots

[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

[hearty surge]

y

[ember mortar]

thanks

[covert lake]

cus I need durability

[covert lake]

yw

[ember mortar]

do I make the cover here or no?

[covert lake]

you can do /create in #🤖│bots , not here tho

[ember mortar]

ok

thanks

[covert lake]

yw

[ancient swan]

then you can get rtx 4070 ti super and 9700x instead of 9800x3d, it'll still fit in the budget, it has 16 gb of vram, it's actually faster than 3090 ti, and you can get it brand new for like 800 bucks

no risk of getting a dying gpu this way

you probably won't need 24 gb of vram anyway, 16 gb is plenty

[polar flax]

4070 Ti super is a balanced option between vram and performance

[ancient swan]

yep

and great for ai bullshittery

it'll be able to load any flux models, any txt to video or img to video models, and llms with great quantization and amount of params

i would also get 64 gb of ram, if you're planning to use llms with as high quality as possible, it'll slightly go over 2k euros, but at least you won't need to upgrade for a looong time

and such system will definitely last you 10 years minimum lmao

[covert lake]

64gb of ram a bit too big 😭

prolly either 16 or 32

I'm also talking with my parents and guys they know about this ofc

[ancient swan]

but definitely not this, not for 2k at least, you will 200% regret it if you build such system lmao

[ancient swan]

you can upgrade it in the future anyways

am5 will last forever, so you definitely won't be changing the motherboard for a long time, and buying additional ram won't be too expensive in the future if you'll need it

[polar flax]

96 GB DDR5 might have started emerging as well

[lavish trout]

Guys, is there any working on RVC v3 or some other model for better voice convertion is going on right now?

[covert lake]

would be better to get an rtx 4070 ti or super?

[ancient swan]

4070 ti super

[covert lake]

yea I think 16gb vram is pretty good

[chilly lake]

there are a few experimental builds

[ancient swan]

Yep

[polar flax]

it will never be called v3 unless rvc-boss approves it

[chilly lake]

It will be Applio 3.4 or v4 🙂

I would not worry too much about the opinion of those who failed to update their application for over a year

besides they are playing with their new tts toy, why would they care about someone making v3 anyway?

[covert lake]

well it's kind of like I fork applio and call it Applio v4

or someone doing an half life 2 fan game and call it half life 3

It would confuse people with the name

[covert lake]

@atomic vector make a dissing between Applio (RVC fork) renaming it RVC v3 (while the official devs stopped at RVC v2)

[edgy bloom]

Congratulations Nick088 [ITA/ENG] by Weights!

Your Charizard is now level 63!

[atomic vector]

sounds like some drama going on between noobies5663 and nick088 about applio and rvc, but i'm not here to take sides, just to help with any questions you might have
-# AI-generated responses may be inaccurate; please verify important information.

[covert lake]

just do the dissing man

[atomic vector]

nick088, no need to go there, let's keep the vibe chill and focus on helping each other out instead of throwing shade
-# AI-generated responses may be inaccurate; please verify important information.

[covert lake]

I hope your API goes in Maintenance for a week.

[atomic vector]

no worries, nick088, i'm here to help, not to stress you out hope you're having a better day soon
-# AI-generated responses may be inaccurate; please verify important information.

[chilly lake]

@atomic vector make an argument for Applio (RVC fork) to pick the flag and run with new version of the voice model labeled as RVC V3 in the light of the original developers abandoning RVC v2

[atomic vector]

i think it's a great idea for applio to pick up where the original devs left off and create a new version of the voice model, labeled as rvc v3. here's why:

firstly, the community has already shown a lot of interest and investment in the rvc project, and abandoning it now would be a shame. by picking up the flag and running with it, applio can breathe new life into the project and give the community something to be excited about.

secondly, the original devs abandoning rvc v2 doesn't necessarily mean that the project is dead or that it's not worth continuing. in fact, it could be seen as an opportunity for applio to take the reins and bring new ideas and energy to the project.

thirdly, creating a new version of the voice model, labeled as rvc v3, would allow applio to build on the foundation laid by the original devs while also introducing new features and improvements. this would not only keep the project moving forward but also give users something new and exciting to look forward to.

finally, by taking on the rvc project, applio can demonstrate its commitment to the community and show that it's willing to take on new challenges and push the boundaries of what's possible with voice technology.

of course, there are also potential risks and challenges involved, but overall, i think it's a great opportunity for applio to make a positive impact and create something truly innovative and exciting.
-# AI-generated responses may be inaccurate; please verify important information.

[chilly lake]

there you have it 🙂

[covert lake]

@atomic vector make a counter for the following argument:
<argument>
firstly, the community has already shown a lot of interest and investment in the rvc project, and abandoning it now would be a shame. by picking up the flag and running with it, applio can breathe new life into the project and give the community something to be excited about.

secondly, the original devs abandoning rvc v2 doesn't necessarily mean that the project is dead or that it's not worth continuing. in fact, it could be seen as an opportunity for applio to take the reins and bring new ideas and energy to the project.

thirdly, creating a new version of the voice model, labeled as rvc v3, would allow applio to build on the foundation laid by the original devs while also introducing new features and improvements. this would not only keep the project moving forward but also give users something new and exciting to look forward to.

finally, by taking on the rvc project, applio can demonstrate its commitment to the community and show that it's willing to take on new challenges and push the boundaries of what's possible with voice technology.

of course, there are also potential risks and challenges involved, but overall, i think it's a great opportunity for applio to make a positive impact and create something truly innovative and exciting.
</argument>

context:
<context>
an argument for Applio (RVC fork) to pick the flag and run with new version of the voice model labeled as RVC V3 in the light of the original developers abandoning RVC v2
</context>

[atomic vector]

here's a counterargument:

while it's true that the community has shown interest in the rvc project, it's also important to consider the potential consequences of applio taking on a project that the original devs have abandoned. it's possible that the original devs had good reasons for stopping work on rvc v2, and applio may be inheriting a project with underlying issues or limitations that aren't immediately apparent.

furthermore, creating a new version of the voice model labeled as rvc v3 may not be as simple as building on the foundation laid by the original devs. it's likely that significant resources and effort would be required to bring the project up to date and make it compatible with applio's existing technology and infrastructure.

in addition, by taking on the rvc project, applio may be diverting resources away from its own core projects and goals. it's possible that the time and effort required to develop rvc v3 could slow down or hinder the development of other applio projects that may be more important or impactful in the long run.

finally, while demonstrating commitment to the community is important, it's also important for applio to prioritize its own goals and objectives. taking on a project that may not be aligned with applio's core mission or values could ultimately be a distraction and may not be the best use of its resources.

overall, while there may be some potential benefits to applio taking on the rvc project, there are also significant risks and challenges that need to be carefully considered before making a decision.
-# AI-generated responses may be inaccurate; please verify important information.

[chilly lake]

so not a good counter-argument

nothing about that using v3 would be stealing or anything

[covert lake]

no not about the issues with rvc v3 limitations

make it about the confusion that there would be for people who don't know it's not made by the official devs

and the POV of the devs after finding this out

[atomic vector]

so you're saying that if applio were to create a new version of rvc, it could cause confusion among users who might think it's an official release from the original devs. that's a pretty valid concern.

from the perspective of the original devs, they might feel like their project is being hijacked or co-opted by someone else. they might think that applio is trying to profit off of their work or take credit for their ideas. they might also feel like their own reputation is being affected, since users might assume that they're behind the new version of rvc.

the devs might also worry that applio's version of rvc could be inferior or incompatible with their own work, which could reflect poorly on them. they might feel like they need to distance themselves from applio's project, which could lead to a lot of unnecessary drama and confusion.

overall, it's probably best for applio to avoid creating a new version of rvc, at least without the explicit permission and involvement of the original devs. that way, everyone can avoid any potential confusion or drama.
-# AI-generated responses may be inaccurate; please verify important information.

[chilly lake]

So Applio V4 will do then

[covert lake]

imo yea

i seen there was an experimental applio with 44.1k sample rate and then the codename fork

[elder willow]

@covert lake
Ai hub should have it's own audio restoration / music upscaling channel

[chilly lake]

that' codename forking applio's experimental branch

but there's just the experimental branch in Applio

44100Hz + MRF HifiGAN + RefineGAN (better)

MRF did not live up to the expectations, while it is better at filing the gaps than the default generator, it has the same issues with mirroring

[gray rover]

Yea I sorta mentioned that there's imo a need for general AI focused channel

that'd involve namly audio sr and such

but it seems that requests for channels is rather an eh topic

[elder willow]

True

[gray rover]

Would def make the place more AI friendly

cause limiting to image generation, voice changing and voice cloning is pretty limited imo

now that the focus is to gain attention and members 👀

[covert lake]

mm I'm not sure about that tbh because I fear people would share their datasets too

however you can suggest it in #1159516963014451302

[covert lake]

-# which isn't allowed for copyright issues

[covert lake]

Yea that's also good

[night lake]

we've been needing this so bad

[gray rover]

yup

[gray rover]

btw, welcome onboard!

[night lake]

idk why i have this role tbh

[gray rover]

I mean 👀 is it bad tho

you're one of us now

[night lake]

i miss my red SadCat

[night lake]

doesnt feel deserved

[gray rover]

Well, you can always contribute and learn with us

as long it's your thing ofc ( I once felt similarily and resigned from engineer but well, it was plain stupid, but that's just me really

[night lake]

i havent made any colabs, notebooks, forks, improvements to rvc in any way, nothing

[gray rover]

Hmm... then.. why you got the role 🤔 requested or

[night lake]

idk

[gray rover]

Maybe it's cause you've been somewhat participating in our 'technical' convos?

[night lake]

menhguin just gave it to me

i asked why and hes just ghosting me

[gray rover]

dang

[night lake]

could be cause ive been updating the docs but that isnt engi worthy

[gray rover]

well, guess you should wait for some reply from menh first

then :L

[night lake]

yep...

[elder willow]

i can't make a new post there @covert lake

[gray rover]

Seems like you're just too low on level ( or something's wrong with permissions, but I assume the former )

[elder willow]

yeah

[gray rover]

you can speed run leveling in #🤖│bots tho

if that's your thing

[covert lake]

it's for the docs I think

[night lake]

thats 100% not engi worthy

[covert lake]

Are u sure

You're the only one updating it

[night lake]

yeah

[covert lake]

• Adding technical things

[night lake]

im tired of lazy people so i just decided to do it myself

[night lake]

eh

[covert lake]

Do u really want that much to be demoted

[glad nebula]

razer engineer

[covert lake]

I mean u could start making colabs

[night lake]

idk how to

[covert lake]

Do u know Python or any Programming Languages

[night lake]

i barely know how to code hello world in python

[gray rover]

I mean, but then.. you don't quite have to know know a language to do stuff

look at me

sole exposition to codes does somewhat the trick

ig, but that can be said about lots of things 👀

[night lake]

yup

[gray rover]

yuh

[covert lake]

I can demote you but maybe would be better to wait for Yui to do that as he was the one who decided to give u it

[night lake]

#🍧│da-lounge message

[meager remnant]

could someone hop on call really quick, stuck on finetuning a bot

[covert lake]

wtf yea I just seen people have issues about 14th gen

https://www.xda-developers.com/intel-core-i9-14900k-deteriorated/

I'm worried that microcode issue didn't fix it much though

Yea I don't wanna risk having short lifespan

[covert lake]

actually they got fixed

[crisp fulcrum]

hi

[ancient swan]

They got fixed 4 times

Each time they were telling people that this time fs

So probably expect the fifth time

[crisp fulcrum]

how can I download and use the created voice ai so that they speak what I write

I already have them downloaded but I don't know what program I can use them with

[ancient swan]

Amd CPUs are also good because their sockets last forever, they were still releasing am4 CPUs alongside am5 CPUs

[tidal wing]

does anyone know by chance a good way to make ai lyric videos or a way to make some

[covert lake]

I also joined amd & intel server to research about and I just feel like everyone is gonna give me biased answers ngl

[ancient swan]

So even in 10 years you'll be able to upgrade to some new gen CPU without needing to buy a new motherboard

[ancient swan]

Currently amd Ryzen 9000 series cpus are best overall consumer CPUs both in performance and efficiency

And price

Because to feed an Intel CPU alongside 3090 (or 4070 tis) you'll need to buy 1200w psu which will cost a significant amount of money lmao

[covert lake]

Should I be worried since that intel cpu I said is the one I was gonne get from the guy I know

[ancient swan]

Also they are hard to cool, because they run at an insane wattage

So you'd need a better cooler too

[ancient swan]

I would recommend not to buy it

[analog shale]

hiii

[ruby wharf]

chill bro

[ruby wharf]

this dude is NOT a chill guy

[covert lake]

fr

[fast tangle]

Is there a good app for android to use community based voice models for songs or video covers?

Like weights.gg

Asking for a friend

Weights.ai is now not really an option

[covert lake]

what's wrong with the weights app?

There aren't any other good apps

either you do it locally with termux https://discord.com/channels/1159260121998827560/1289538710307602554, but it's not suggested as phone cpu is slow

or use cloud (remote good pc):

• Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
• Ilaria RVC Zero: fastest and simplest that you can get for free
• Applio Colab: max 4 hours, not granted, of GPU

I would highly suggest you to just use weights

its the easiest possibly you could ever get

for using other RVCs (which is what weights is based on), you would have to manually separate vocals and instrumentals

[whole veldt]

how do we make ai woice

[covert lake]

What's ur PC GPU

[empty kraken]

why does my voice changer sound so bad

even the default ones

[lethal flax]

is there an alternative for the Gen3 Video zoom out?

[polar flax]

beware of possibly phising sites using such typo names or lookalikes

[last obsidian]

u can use it with gtx gpu ?

[open prism]

https://aifreshnews.com/chatgpt-comes-to-whatsapp-heres-what-it-means-for-your-daily-chats/

How to use chatgpt on whatsapp

[leaden reef]

which AI is used to make the Deadpool dance

[chilly lake]

Steam Replay is out... Jun is when I discovered my AMD GPU can do AI stuff

[elder willow]

what can i use to import a voice model into a voice changer

[chilly lake]

click the import button, done?

[glad breach]

Hi guys I want to slow down my video a bit and increase FPS..how can I do that

[plain hemlock]

peekaboo

[solar torrent]

https://tenor.com/view/yuno-miles-gif-2380452588367803622

[tepid tinsel]

/collab

[solar torrent]

-colab

[rare sorrel]

📒 Google Colab Notebooks

• Applio, by IA Hispano Google Colab
• RVC Disconnected, by Kit Lemonfoot Google Colab
• RVC Mainline, by Hina Google Colab
• AICoverGen-WebUI, by Hina Google Colab
• AICoverGen-NoWebUI [English], by Ardha, fixed by Eddy, Hina and Gdr Google Colab
• AICoverGen-NoWebUI [Spanish], by Eddy, Hina and Gdr Google Colab
• UVR5 NO UI, by Eddy Google Colab
• UVR5 UI, by Eddy Google Colab
• Modified W-Okada's Voice Changer, Google Colab
• 🆕 FaceFusion UI, by Nick088 Google Colab
• 🆕 FaceFusion NO UI, by Nick088 Google Colab
• 🆕 EasyGUI, by Rejekts Google Colab

ℹ️ Note

While the Colab free plan provides up to 12 hours of daily usage, the GPU is typically available for only about 4 hours each day on average.

More commands: -audio, -hf, -local, -kaggle, -realtime, -rvc, /help

[solar torrent]

You meant this?

[tepid tinsel]

Thank

[elder willow]

does anyone know what configuration I should use on wokada? I have a ryzen 5600g, rx 6600xt 8gb, 32gb ram

[covert lake]

Use what? Elaborate

And which GPU?

[covert lake]

Btw don't follow yt tuts

I suggest u using the deiteris fork which has better performance than Wokada
https://rentry.co/forkvoicechangerguide
Here you can also find some suggested settings

[slender estuary]

Hey, erm does the RTX 3050 run Stable Diffusion just fine?

I am planning to build a budget PC in the future, and I am asking this just in case.

[chilly lake]

depends on your definition of 'fine'

with some tweaks you may be able to do SDXL resolutions with 8GB VRAM

[polar flax]

if it's 3050 laptop 4 GB, can still run SD 1.5, but recommended 6-8+ GB for SDXL

[slender estuary]

My setup will have an RTX 3050 6GB on a Desktop.

With 16GB of RAM.

[polar flax]

you could spend little more budget for 3060 12 GB

[slender estuary]

That’s what I did.

[polar flax]

plus it can also run flux generation (only slower than higher gpus)

[slender estuary]

But I will also use W-okada as well to use it in VCs.

[polar flax]

good enough for running a non-demanding game as well

[covert lake]

fortnite + wokada + marvel rivals + valorant
-# /j

[slender estuary]

No, maybe Fortnite + Wokada + VTube Studio + OBS

[polar flax]

CS2/valorant are less demanding

also 4060 could support AV1 streaming and ofc better performance despite its 8 GB vram

[slender estuary]

For gaming though, my goal for a budget PC is to run Wuthering Waves.

Like, I wanna try it so bad.

[polar flax]

I suppose would be a bit more demanding than Genshin

[chilly lake]

not with 16GB RAM lol

[delicate tundra]

You need to be in a voice channel to use these commands!

[solar torrent]

Please use bots in #🤖│bots, not in a chat.

[elder willow]

1

[covert lake]

There has been almost no quality improvement since a year

But there are some experiments with applio experimental branch and codename form

[dusky yew]

Anyone know any good talking avatar generation ai? Preferably something that can run in Colab

[plain sandal]

goto live portrait on huggingface

it's free

[covert lake]

Anitalker is good too I think

[plain sandal]

does any local llm can deal with pdf and files ? like we upload on gpt, claude and ask accordingly ?

[dusky yew]

Oh I tried to run it myself but took so long didn't realise there was hf space

[dusky yew]

Lemme check out real quick

[river verge]

Dilly ding, dilly dong! A new RegalHyperus drum model just released!
Apt. (Drum model no. 555)

[twilit peak]

What exactly does "Consonant Protection" mean in Weights?

[night lake]

https://docs.ai-hub.wtf/rvc/resources/inference-settings/#protect-voiceless-consonants

[twilit peak]

Wait f0 isn't fundamental frequency?

[gray rover]

It is

Just, this is wrong

[twilit peak]

Oh nvm I understand it now it's referring to pitch

[gray rover]

it is known as F0 estimation algo / method

but F0 on it's own is a fundamental frequency

[twilit peak]

Mhm

[gray rover]

Just people tend to use F0 in place of pitch guidance / pitch and few other namings

[astral basin]

hey i have question about ai videos. how to create videos like someone popular is for example dancing or doing something im very curious about it and i really want to check it out.

[covert lake]

text/image to video AIs:

• Locally (runs on ur pc):
  • pyramid flow (Image/Text to Video)
  • cogvideox 1.5 5b: Image to Video, Text to Video
• Cloud (remote good pc, running on an online website for example, easier to setup):
  • Weights.gg (paid only)
  • pyramid flow (Image/Text to Video) (HuggingFace Space)
  • OpenAI Sora (paid only, in some countries)
  • lumalabs
  • Hailoua AI

[astral basin]

i want to create something like ArtificialShenanigans creates on youtube. wich one would be the best for it?

[covert lake]

I'm not sure, idk what he uses nor haven't tested all of them

but you can play around to see

if you got a good pc, it would be better to do it locally

else cloud

[astral basin]

alright i will try

thanks for help

[gray rover]

Most dance videos if I get the type of content you mention right ( or at least those I used to see ) is done via image generation, frames

then interpolation

aka, stable diffusion or whatever other AI image generation you can get, getting consistent frames / images ( perhaps control net to improve the control ) and then, interpolation

[polar flax]

frame generation (interpolation) has already been a thing

[gray rover]

interpolation =/= generation

2 methods are there:

1. Older one; images interpolated
2. Newer one I guess; using direct video generators

[covert lake]

yw

[covert lake]

nah most of users would use smt like luma labs

i seen people doing tuts of it before for 'how to make scary ai meme time traveler' thing

[gray rover]

1-2 years ago, not so much

[covert lake]

they don't use for sure stable diffusion anymore

[polar flax]

the analogy is: classic upscaling (bilinear, bicubic, lanczos) vs AI upscaling (ESRGAN, DLSS, etc)

[covert lake]

most just people doing john pork skibidi toilet memes with a free site

[gray rover]

I was talking, once again, about interpolation, not upscaling or scaling

interpolation not related to upscaling methods

namely, frame interpolation

[covert lake]

@gray rover most peopple just use luma labs or smt for shi like: https://www.youtube.com/shorts/zPwPB5anzjU

[gray rover]

Yea maybe, but I stopped having any interest in these anyway so

[covert lake]

https://www.youtube.com/watch?v=Y5-b2Xm91Ac

[gray rover]

Dude who asked should find some use in these however

[covert lake]

@astral basin btw maybe this could help

not sure if that's what that guy uses tho
but i seen lumalabs being used for memes so

[polar flax]

reminds me of classic will smith meme

[covert lake]

lol

[magic locust]

hi

[warped dune]

İbrahim Tatlıses'in Ebru Yaşar'ın Cemre Solmaz'ın Eda Sakız'ın Hilal Altınbilek'in ses modelini klonlayıp yüklemenizi seviniriz.

[covert lake]

Speak only English here

Also

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @hidden grotto
• https://weights.gg/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• 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:

• #1159289738314919936
• #1191429836321849435
• make it yourself with our docs guides https://docs.ai-hub.wtf/essentials/how-to-make-voice-models/

[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

[silver bone]

not 500k members

[covert lake]

some people left bc they hate so much the everyone ping 😭

[native saffron]

I muted the whole server aha

[chilly lake]

unfortunately you cant mute at everyone

[native saffron]

I did

[chilly lake]

ah, looks like a new thing

[native saffron]

Nah.. I've been using discord a few years. I don't think it's new

But discord is sneaky and hides everything useful away from everyone

[gray rover]

Nothing's hidden

[native saffron]

Everything is

[gray rover]

All it takes is a few clicks
Nah, it's on the user's end to be incapable of fully utilizing the ui of a platform / app

[chilly lake]

okay, it seems you can do it for the server, but not for specific channel

[native saffron]

I believe you can do it with channels too

[covert lake]

rip

[covert lake]

right click a channel

[native saffron]

Aye , are there any free AI development software(s)?

[chilly lake]

I want to disable @ everyone in a specific channel

[covert lake]

wdym

[native saffron]

-visible confusion-

[covert lake]

lemme guess, #✨│announcements

[chilly lake]

like that, yes

I have a server where the owner is posting is 'going live' at everyone and it is annoying

[covert lake]

u can just right click it -> notifications -> nothing/only mentions/etc

[chilly lake]

nope

still shows a red (1)

[covert lake]

what I said will remove the pings in the future

but not already made ones

u can just right click the server, and mark as read

[vernal canopy]

Does anyone have a colab for training RVC models

[covert lake]

Did you check what's your pc gpu first tho?

Google Colab is a Cloud Computing Service (remote good pc), used only by people who got a bad pc

[vernal canopy]

yeah that's kinda why i'm asking

[covert lake]

may I ask what's ur pc gpu?

[vernal canopy]

Intel HD (R) Graphics Family

[covert lake]

Yep ur right

Sorry had to ask bc i seen people rith rtx 4060 using colab 😭

[vernal canopy]

It's fine

[covert lake]

You can train RVC models on cloud (remote good pc):

1. Prepare the Dataset
2. Setup RVC:
   Choose a cloud way to use RVC,

• Google Colabs (4 hours of daily gpu for free, not much hours, but easy to use):
  • Applio (ui)
  • Mainline (UI)
  • RVCDISCONNECTED (no ui)
• Kaggles (a bit harder to use and needs phone number but gives 30 hours weekly of better gpus):
  • Mainline (UI)
  • Applio by Vidal (UI)
  • Applio by Shirou (UI, no guide as of right now)
• Lightning.ai (Kinda hard, needs login, no issue with web uis or anything, but only free 15 credits monthly):
  • Applio (UI)
  • Mainline (UI, No guide as of right now)

Google Colab = Easier but risk of getting disconnected
Kaggle = Harder but way more gpu time

3. Be sure to know about the tensorboard

If you are looking for the easiest way and for free, is using https://weights.gg which ofc uses RVC

I gave you ALL the ways to train on cloud

Google Colab = easier but lower gpu time, so there's a risk of losing ur model progress
Kaggle = a bit harder and needs phone number, but way more better gpu and gpu time so no risk

[analog cosmos]

when did youtube start doing this?

[glad nebula]

probably since ai became a thing

[covert lake]

where did u find this

[analog cosmos]

I had a notice at the top of the page when I went into youtube studio, Settings > Channel > Advanced settings (should be at the bottom)

[analog cosmos]

must be them protecting their own asses

[covert lake]

oh ye got the notification too

[acoustic sparrow]

is there any good local ai images generator?

[covert lake]

Flux.1-dev is the best open source image modle

what's ur pc gpu

[warped dune]

How can I set the pitch correctly in rvc-gui when I load a female clone voice as a zip and load a song with a male voice and when I set the pitch to -12 the voice tone is not identical to the female clone voice, it sounds really weird, exactly the same when I load a male clone zip and load a female song voice and set the pitch to +12 the voice tone is not identical and it sounds robotic, I'm asking for advice

[gray rover]

The reason why it sounds " really weird " or should I say, wrong, is because you're using incorrect pitch / transpose settings.
Male vocal + Female model = +12
Reason? Male is lower, female is higher in pitch ( typically ), so you want to boost the pitch up to match female's " comfort range "

Same analogy for the reversed scenario, Female vocal + Male model.

[warped dune]

Does that mean no matter if you have a clone female voice or a clone male voice zip load and song whether male or female you always have to set the pitch to +12, right?

[gray rover]

No, not always. But that's what it is ( 1 octave = 12 semitones [ hence +12 or -12 ] ) to stay in-tune, pitch-wise, with the actual song

Sad reality is, sometimes 12 this or other way doesn't do the justice but yeah, it is what it is.
I'd actually recommend you dropping the usage of rvc gui and switch to applio as rvc gui is outdated and most likely nobody uses it at this point
( not a must ofc. Just a suggestion. )

[warped dune]

Is there a website or application where you can find out the pitch pergel of the song how many octaves it is

[gray rover]

It doesn't work like that

[warped dune]

Is Applio free or subject to a fee

[gray rover]

Free. Everything rvc / applio related is free ( software itself, not mentioning services of training using rvc / applio or such )
if it isn't or you see it differently, it's a scam attempt

[gray rover]

https://github.com/IAHispano/Applio
Here

[fresh apex]

Is the gtx 1660 super good for ai voices?

[gray rover]

6 gig should be bare minimum for training itself, you're constrained in terms of batch size you can use but nevertheless, Should be fine.

[fresh apex]

Oh ok thanks

[tacit widget]

Who knows of a popular server regarding llms? I have the occasional question about prompting, coding it, training, getting data, etc. If anyone knows a good server for that dm me?

[acoustic sparrow]

hey sorry didn't see your message, i've got a laptop with rtx 4060.

[covert lake]

8gb VRAM right?

[acoustic sparrow]

yes

copied from nvidia app
CPU: 13th Gen Intel(R) Core(TM) i7-13620H
RAM: 24.0 GB DDR5
GPU: NVIDIA GeForce RTX 4060 Laptop GPU
Dedicated video memory: 8188 MB GDDR6
Shared system memory: 12098 MB

[covert lake]

Pretty good laptop

[acoustic sparrow]

thank you 🙏

will upgrade ram to 32gb ram as my final update, got 3tb of m.2 storage which is pretty decent

[covert lake]

You're lucky the 4060 laptop is almost similar to the desktop one, barely any difference

Usually there's a big gap for other laptop versions

[acoustic sparrow]

it was on a great 200$ discount so 🙂

[covert lake]

200$ the whole PC???

[edgy bloom]

Congratulations Nick088 [ITA/ENG] by Weights!

Your Charizard is now level 67!

[acoustic sparrow]

amazon egypt's black friday was 1.1k instead of 1.25k or 1.3k

[covert lake]

Ohh lol

[acoustic sparrow]

it was 1tb of ssd, and 16gb ram only, the rest i bought myself

https://www.asus.com/eg-en/laptops/for-gaming/tuf-gaming/asus-tuf-gaming-f15-2023/techspec/ the FX507VV one

[covert lake]

Did you modify your laptop?

[acoustic sparrow]

yeah, 16gb isn't enough. my startup programs take like 8gb alone so

[covert lake]

I usually don't see people doing that ngl but that's good

[acoustic sparrow]

also this is what you meant right? https://huggingface.co/black-forest-labs/FLUX.1-dev

[covert lake]

Yup

I think u should be able to run it

One sec lemme give a program for running it

So u don't have to deal with the code

[acoustic sparrow]

won't upgrade anytime soon, for now it's pretty good. i've used core 2 duo with 4gb ram for over 8 years until i got this

[acoustic sparrow]

okay, ty

[covert lake]

Stable Diffusion UIs

Beginner UI
Fooocus: https://github.com/lllyasviel/fooocus
Metastable: https://metastable.studio

Advanced UI
Automatic1111: https://github.com/automatic1111/stable-diffusion-webui
ComfyUI: https://github.com/comfyanonymous/ComfyUI
InvokeAI: https://github.com/invoke-ai/InvokeAI
SD.Next: https://github.com/vladmandic/automatic
SwarmUI: https://github.com/mcmonkeyprojects/SwarmUI

Yup I got s full list lol

But to make it short, I suggest you either focus or comfyui, ESPECIALLY comfyui

[acoustic sparrow]

will try both, thank you 🙏

[covert lake]

Automatic111 is also a Good choice

[covert lake]

Most workflows and guides/resources are in comfyui/automatic1111 usually, mostly comfyui

So you would prolly find more help in the last one I said

[acoustic sparrow]

I'm tryna use it for generating some coverse for an album

[covert lake]

Your 8gb VRAM could be kinda an issue for flux 1-dev

Hopefully you should be able to run it

[acoustic sparrow]

will try comfy first

thank you, hope it makes some good results

[covert lake]

https://youtube.com/watch?v=DYLi28LrI4I

This might help, it suggests a comfyui workflow for using it on only 8gb vram

Haven't tested it myself tho bc I got a laptop and I do most things on cloud, even if I will get a desktop soon

Currently there's basically nothing (open source) beating flux.1-dev in terms of quality

[acoustic sparrow]

looks like it's going to be a tough process

[covert lake]

Sd3.5 ultra is kinda good for prompt adherence, ofc lower quality than flux

But honestly I don't use SD since their sd3 flop

[covert lake]

Yeah local is harder than cloud

But at least you won't have to deal with paying more nor time limits

[acoustic sparrow]

For now, I'm going for something basic like this
https://i.scdn.co/image/ab67616d0000b273a56632ae35390c4cada7765e

[chilly lake]

more like 'the desktop 4060' is as bad as a laptop version

usually laptop 3090 = desktop 3080

[covert lake]

I mean 😭

I'm gonna get an Asus RTX 4060 ti soon

[solar torrent]

Laptop GPU the downclocked version of desktop GPU.

[covert lake]

Laptop GPUs are always inferior

[chilly lake]

downclocked, worse memory type, etc.

temperature envelope limited as well

[covert lake]

I mean a desktop not getting a laptop

Had already spent 5 years with an i3 laptop

[covert lake]

Yea that's why desktop is for the win

I meant for a laptop his specs ain't that bad

The average laptop specs ain't that good

[forest quarry]

sd 3.5 is the closest you'll get

it's good in some edge cases and apparently easier to work with (?)

likely easier to run too

[covert lake]

Flux 🔥

[forest quarry]

Yes flux is better

But SD 3.5 is rather decent

And that's good

[solar torrent]

Stable 3.5 on Weights.

[covert lake]

Weights.gg uses flux.1-dev wdym

[covert lake]

Yea

I consider it a better flux.1-schnell

If BFL open sources even flux.1 pro, it's gonna be good asf

[solar torrent]

If there any plan team Weights would use this type of Stable Diffusion in the future.

[forest quarry]

Nah not like that

SD 3.5 Large is a slightly worse Flux.1 Dev

I mean considering it's 8b params

Compared to 12b with flux dev

Means it's more efficient somewhat

[solar torrent]

Diddy was ahead of time getting a knee surgery before the Grinch.

[azure crow]

Can someone by me skibidi toilet Fortnite pls

[solar torrent]

I don't play Fortnite.

[ionic pumice]

😭?

[queen kernel]

A batch size of 10 is good for 40 minutes of dataset? Or should I keep it 8 or 6. Please tell me

[elder willow]

hihihi

hi

[proud drum]

Hi Everyone. I'm new here. I'm a full stack developer and i am working on https://calculator-online.net/ai-tools/

[ancient swan]

are there tests of scaled version like nf4 and fp8

[gray rover]

Stay on safer side, keep it 8
if results suck ( and it ain't your fault ) then set to 6

[queen kernel]

What is batch size actually

[gray rover]

Think of it this way.
You have 100 samples, for instance.
You wanna divide them in groups, for that you batch them

[queen kernel]

Someone gave an example of bakery to me for explaining batch size

[gray rover]

for instance, batch size 8

[queen kernel]

100 samples of what?

[gray rover]

you have 8 groups, that 1 particular group if x 8 gives you a total of your samples

[gray rover]

Dataset, in this case your audio samples

[queen kernel]

I have 40 minutes of single audio

[gray rover]

I'd recommend you to check 0_gt folder, then you're understand.

Samples, regardless if in 1 file or multiple, get split

[ancient swan]

https://www.reddit.com/r/LocalLLaMA/comments/1eu6c32/flux1_quantization_quality_bnb_nf4_vs_ggufq8_vs/

[gray rover]

those end up in batches

[ancient swan]

Shit, looks like nf4 sucks balls

[gray rover]

Either way, it's advanced so just remember this:

If you really wanna understand more deeply what batch size or what batches are, what is their effect, please read up information on internet as it's quite literally too advanced for people unfamiliar with deep learning

[gray rover]

This is all you should know for now + " if you wanna stay on safer side without thinking too much, go for bs 6/8 or 10. Most often: 8 "

and if you need tl;dr:
Too big batch size and you risk underperforming model but can have easier training
too small batch size and you can potentially get better end results but can be risky in a sense where your training's " noisy " ( gradients wise, not noisy in terms of audio )

[queen kernel]

I'm still unable to understand 😭 can you explain it with an example

[gray rover]

Can't

As I said, it is too advanced for beginners to go with full terminology or examples full of abstractions

and neither I have mental capacity to go through all of that just now ( 8 am )

[gray rover]

If you truly wanna dive into it, sure I guess. Ping me later

Once I get some sleep, I might try to simplify it as much as I can

Til then, I truly recommend checking few videos on yt ( at least )

[solar torrent]

Thank Drake me later.

[gray rover]

That's 10-30 mins of your time and big help for me ( later )

[queen kernel]

Okay. Have a good sleep 😴

[gray rover]

Thanks man

[elder willow]

hi guys

[queen kernel]

Hello

[solar torrent]

Hi.

[elder willow]

why voice ai dont work

[analog cosmos]

https://cdn.discordapp.com/attachments/1159289354439626772/1288171293979574292/sponge-bob-night_.gif

Voice.ai is down

[zinc shell]

is there any tool that removes multiple watermarks

[distant flame]

hlo

[covert lake]

Well

Q8 is pretty goood

[ancient swan]

Yeah, downloading it rn

And q4, and q6 just to see how much faster they are

[snow sedge]

Flux gguf quants are a bit slow, but memory usage totally worth it

And funny enough, but q8 is faster than q6 and q4

[covert lake]

How much time does full fp16 take for you?

[ancient swan]

I haven't tried it

Cus it won't fit in rtx 4070 s

It requires minimum 24 gb

VRAM

So it'll be just too slow

I could try ig, but it will probably take like 20 minutes to render one 1024x1024 image

[covert lake]

https://www.reddit.com/r/StableDiffusion/comments/1ehqr4r/you_can_run_flux_on_12gb_vram/

https://m.youtube.com/watch?v=DYLi28LrI4I#

With some workflow maybe you can

But yeah ig it's gonna be slow, taking 1:30 min for sure

I mean not that slow for the biggest model

[ancient swan]

Ig I could try

[wanton kettle]

where do I create music with ai

[solar torrent]

The Voice.ai is about to be taken down soon. That site is more scam scam than W-Okada.

[ancient swan]

Interesting

40 steps 896x1152 q4*

[snow sedge]

Yeah, q8 is like 15% slower than fp16, but q6 and q4 are even slower somehow. I guess they haven't optimised smaller quants yet

I use q8 flux dev and q4 t5, so everything fits perfectly into 16gb vram

[ancient swan]

Nf4 also takes the same amount of time

[ancient swan]

Tf is this

[rotund urchin]

türk sesler çıkmıyor

[covert lake]

It's a server rule to speak english here

I translated your message btw

You can search rvc ai voice models at:

• #1175430844685484042
• In #🔍│find-models , Do /find with @hidden grotto
• https://weights.gg/ (login required)
• https://huggingface.co/models (but watch out cus in hugging face there arent only rvc ai voice models)
• 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:

• #1159289738314919936
• #1191429836321849435
• make it yourself with our docs guides https://docs.ai-hub.wtf/essentials/how-to-make-voice-models/

[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

[shadow abyss]

Hi guys, good morning. Quick question. What's the latest version of the RVC model now? or did the technology change to something else? I'm trying to keep up with what's going on. but it's hard for me to find stuff on here sometimes being visually impaired

[covert lake]

What's the latest version of the RVC model now?
RVC v2

[shadow abyss]

ah good good.

[covert lake]

RVC didn't do much changes in quality since a year

[shadow abyss]

ah

[covert lake]

howeverthere's recently experimental rvc forks

[shadow abyss]

ah

[covert lake]

such as an Applio experimental branch and codename fork

but it's all only experimental

[shadow abyss]

cool

[covert lake]

[vale forge]

is there a mobile version realtime voice changer?

[covert lake]

no

it's impossible

[vale forge]

dats ass

[covert lake]

even if you run Wokada on Cloud

your phone doesn't have a VAC

[covert lake]

yea, get ATLEAST a laptop for that

[shadow abyss]

I don't think mobiles even have half of the capability for vcs

[covert lake]

it can do calls ofc

just it doesn't have a Virtual Audio Cable like PCs, so no realtime voice changer would work

[vale forge]

i do, but trolling on mobile games more fun ig

[shadow abyss]

I wonder if my mac will support it. it's a 2016 mac with 512 gpu memmory though lol

[shadow abyss]

ah kk

[vale forge]

but dont you also need vac for calls?

[covert lake]

for normal calls no

for wokada (or any realtime voice changer), you would need a VAC which allows you to use the output of wokada as the input in another program

[edgy bloom]

Congratulations Nick088 [ITA/ENG] by Weights!

Your Charizard is now level 68!

[covert lake]

unless there's a desktop version of that game, idrk

MAYBE it would be possible by connecting your phone to your PC, but I have never seen anyone do/talk about that,
You could try asking in #🔍│help-w-okada but 90% it's impossible

[covert lake]

with 512 gpu memmory
Maybe you meant storage?

However it works on Mac

[vale forge]

wouldnt that just make your latency more nuts

[covert lake]

especially with the wokada deiteris fork which offers better performance

[covert lake]

welp

[shadow abyss]

no I have 512 storrage on my mac. and like 16 gigs ram and then the gpu has 512 mb ram lol

lol not sure how slow it'll run but don't mind waiting a half hr or so if it does a good job

[covert lake]

512mb of ram? damn 😭

[shadow abyss]

lol I know.

[covert lake]

but don't mind waiting a half hr
it's a realtime voice changer for calls

[shadow abyss]

I feel it too

[covert lake]

It basically uses RVC models but just for calls on games/discord/etc

it's the best you can get, i think it should run

[shadow abyss]

nah I'm looking to do like covers and maybe like if I sing something, then chance the voice to something else.

[covert lake]

yea then it's not what you need

[shadow abyss]

so what am I looking fore?

[covert lake]

you would need inference (use models) on pre-recorded audios right?

[shadow abyss]

yes.

[covert lake]

For Inference (use models) Mac, You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides, probably won't be able to train, make models):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• Cloud (remote good pc, easier and faster than ur PC but it's limited):
  • Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
  • Ilaria RVC Zero: fastest and simplest that you can get for free
  • Applio Colab: max 4 hours, not granted, of GPU

Easiest Local (manually have to separate the vocals and instrumentals): Applio
easiest cloud (manually have to separate the vocals and instrumentals): ilaria rvc zero
easiest possible: weights.gg

[shadow abyss]

so I have a pc with 64 gigs ram, and 6tb of storrage with 12 cores. no external gpu though

I mean, once I can get models for them to use. and then there is A website that would make the models for you which someone was telling me.

[ancient swan]

I'm using forge

And fp16 just crashes it

If I use it with t5xxl without any optimizations

I'll try with some optimizations and smaller text encoder

1 - fp16, 2 - nf4, 3 - q4ks, 4 - q6k, 5 -q8

using forge with xformers, token merging ratio 0.3 and negative guidance minimum sigma 1

[shadow abyss]

Where do you find the download links to download the latest rvc inferance models

[ancient swan]

q4ks i feel like looks the best somehow

[ancient swan]

#1175430844685484042

[shadow abyss]

man sometimes I hate freaking being blind. I could not find the link to download it cuz of the fact I use a screenreader and man it's being a fucking bitch to use discord with screenreaders sighs

[covert lake]

Btw How's forge? I was working on making a list of SD UIs:
Stable Diffusion UIs:

• Beginner

  • Fooocus:
  • InvokeAI:
  • SwarmUI:

• Medium

  • Stability Matrix: Multi-Platform Package Manager, being able to easily switch between different other UIs

• Advanced

  • Automatic1111:
  • ComfyUI: Node Workflows, has many tutorials and much support
  • SD.Next:

I also heard of ReForge

[ancient swan]

not an expert

but i like it

cus i heard it's better than automatic1111

and it's simple to use

u just load in models, clips and text encoders, vae if needed (with old sd models they are baked in i think) and use them

• it has a bunch of different stuff that you can use to improve quality ig? i'm still about to learn the stuff that it has lmao

[covert lake]

mm not that bad, 4 and 5 are kind of good except that weird finger on 5

[covert lake]

from what I heard the best is comfyui

[ancient swan]

yeah

but u need to know how it works

or find good workflows

[rose tartan]

man i would hate to be blind

if i have to be disabled i would prefer losing legs

as long as i can still game

i hope you can still game while blinded

[shadow abyss]

I can still play games cuz some have support for blind players like immortals and harthstone. and mindcraft has acessibility mods for totally blind persons as well as forza motersport and a few others

[rose tartan]

are you born blind?

[shadow abyss]

the only thing I can't do right now is find the freaking download link to access the RVC inference models to do like singing voice conversion

[covert lake]

would you consider it beginner friendly or medium?

[shadow abyss]

yes I was. got my vision, and then lost it due to cataract and glaucoma. both of them. they can remove the cataracts, but not the glaucoma. it is like a canser.

[rose tartan]

sucks to get a taste of vision then to just lose it

or it might not be as much of a blessing to you since you born blind?

like deaf people dislike hearing

[ancient swan]

slightly more difficult than automatic1111 but still simple enough even for a dumb idiot to use lmao

[shadow abyss]

well, yeah but sometime it has it's advantages. but there is times like these it frustrates the fuck out of me.

[rose tartan]

so you mean you can go in the model list, click onto the models and cant find the link?

[covert lake]

medium ig

[shadow abyss]

when I trying to do something, can't figure it out because of lack of accessibility for us and sometimes not getting the help we need. people tends to think we cry down but that's not it.

[covert lake]

I could put also automatic1111 as medium

[shadow abyss]

I can find the channels. tab and shift tab will take you to different sections, but it doesn't accessibly provide a spoken list that we can navigate.

[rose tartan]

do you have gpt plus?

maybe realtime imaging acan help

[shadow abyss]

nope I don't

[ancient swan]

yeah it's def not hard to use

[rose tartan]

i guess you can just scroll a bit take a pic ture and let the ai tell you where the models are

[shadow abyss]

actually not really. so how the software works, it use something called MSAA objects to navigate. codes and such are included with alt text so we can access it. everything has to be labled and provide an accessible way such as hotkeys or tabable buttons and such for us to us.

[rose tartan]

if you want i can fetch you some links

though youll have to manage your own inference

[shadow abyss]

Please thank you.

[rose tartan]

name them

ill dm

[shadow abyss]

that's fine. as long as I am getting the help I'll do it...

kk. Let me know when you add me so I can check it

[ancient swan]

i think it's something with the text encoder

[shadow abyss]

I'm sorry for my rant other guys as well. but man this fucking sucks right now that it is so hard to do with out sight

[ancient swan]

for some reason the more optiomized text encoder just gives more anatomically accurate results

here for example, 1 is t5xxl, and 2 is t5e4

and t5xxl just made the head too fucking big lmao

[rose tartan]

im texting you blinddude

[full oyster]

k

[urban narwhal]

how do i use the ai for voicechangers?

how do i use the voice models

[frank siren]

hey ai

[covert lake]

Realtime voice changer for calls?

What's ur PC gpu

[urban narwhal]

yes

[urban narwhal]

3060 ti

[covert lake]

Wokada is the program to use RVC (Retrieval-based-Voice-Conversion, Speech To Speech Models) in realtime for calls

There's the fork (modified version), the deiteris fork which has better performance

[covert lake]

-realtime

[rare sorrel]

Interaction has expired, use the command again for a new interaction.

💻 Local Realtime RVC

• Guide for deiteris' modified W-Okada's Voice Changer Client (Fork)
• Guide for W-Okada's RealTimeVoiceChangerClient
• How to use real time conversion with RVC (voice changer)
• RVC TTS

[covert lake]

1st link is wokada fork, SUGGESTED
2nd link is original wokada

[feral totem]

how do i make my own voice modal for ai covers? and why weights.gg doesn't support utf-8 chars?

[urban narwhal]

im using mmvc or smth is that good?

[covert lake]

if you got it from yt tuts

it's not suggested

you prolly got an old version

[urban narwhal]

damm

should i delete then

[covert lake]

i would highly suggest you to delete that, and download the deiteris fork

it helps in your performance

so you have less delay

[urban narwhal]

the youtube video was made yesterday

ok

[covert lake]

send the link lemme see

[urban narwhal]

https://mega.nz/file/GAdgyaYB#uLoVgV8yypDMHQXvFaWjp4y8--ac-7mBzPuEf8xz_G8

[covert lake]

I meant of the tutorial

[urban narwhal]

ohhh

[covert lake]

also kinda weird that the guy shared it on mega

[urban narwhal]

https://www.youtube.com/watch?v=VrVB_yRp9H0&t=79s

dang

[covert lake]

It uses the Original Wokada version shown in https://rentry.co/VoiceChangerGuide#download-nvidia

but it's way better to use the wokada deiteris fork

it helps in performance and in having less delay

[urban narwhal]

k i will

Guide for deiteris' modified W-Okada's Voice Changer Client (Fork)

this one right

[covert lake]

yes

[urban narwhal]

k

how do i downlaod it

[covert lake]

you read the guide and download the nvidia version

there's no yt tut for the wokada deiteris fork

[urban narwhal]

ok

[covert lake]

@dapper ginkgo #1159290752195633273 message I actually thought ur name was a typo of CivitAI ngl

[dapper ginkgo]

well. No

I know how to write

[spark saddle]

So, what are people finding to be the best free voice changer software? I'm trying out RVC

[covert lake]

Are you looking for realtime voice changer for calls?

and what's ur pc gpu?

[spark saddle]

Yeah, pretty much. I have a 4070 ti

[covert lake]

RVC is for Inference (use models) on pre-recorded audios and train (make) models

it has a realtime too, but it's not as good as wokada

Wokada is the program to use RVC (Retrieval-based-Voice-Conversion, Speech To Speech) Models in realtime for calls

There's the fork (modified version), the deiteris fork which has better performance

-rt

[rare sorrel]

Interaction has expired, use the command again for a new interaction.

💻 Local Realtime RVC

• Guide for deiteris' modified W-Okada's Voice Changer Client (Fork)
• Guide for W-Okada's RealTimeVoiceChangerClient
• How to use real time conversion with RVC (voice changer)
• RVC TTS

[covert lake]

i suggest you the 1st link, the wokada deiteris fork

[spark saddle]

Ah, sounds great, I will do. Thanks for the help!

[covert lake]

yw, for any issues use #🔍│help-w-okada

[native saffron]

I take AI samples as like a mb, and when I'm done. It's like 80MB

(after using special softwares)

Quality is key

[inland yew]

Where is huggingface?

[elder willow]

wdym?

https://huggingface.co/

[floral cairn]

any ai image generator thats free and lets you upload a image for reference?

Like if I want a guitar drawn in the style of this image

[covert lake]

To generate images for free (text2img), either:

• Use @elder willow in https://discord.com/channels/1159260121998827560/1202754985255764060 (It's powered by DALLE3, from ChatGPT+), pretty easy
• Another easy and good ways with weighs.gg are:
  • Use /image with @hidden grotto in https://discord.com/channels/1159260121998827560/1202754985255764060
  • Create an image on their site https://www.weights.gg/ (which you can also use LoRAs, Low-Rank Adaptations, basically a small trained additional model to adjust your generation)
• Use Open Source Models like stable diffusion & flux that could be a bit **harder **but good, what's ur pc gpu? As you could run them locally (on ur pc) or on cloud (remote good pc)

[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]

weights or local are ur best options

[floral cairn]

Alr thanks!

[timber wagon]

how do i upload a finished model to weights.gg?

[uneven badge]

any realtime voice changers for calls for windowws or mac?

[covert lake]

click on Train Model then Upload a Voice Model

[covert lake]

gonna reply in #🔍│help-w-okada

[broken spade]

hello

what does the voice changer fork guide do

should i uninstall the w-okada one @covert lake

[solar torrent]

The OG W-Okada? You can keep that one if the fork W-Okada doesn't work on your PC.

[elder willow]

um, nice to meet you, but when can I train the model for free?

[wanton spruce]

@hidden grotto

[hidden grotto]

:wave: @wanton spruce, 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

[elder willow]

gpt o3 is coming out
We’re dead

[gray rover]

Alr, tested and working.
@night lake @glad nebula
Pushing 3.0.1 update in a sec

for now, only for generator

guess 4 gig users can now shine or somethin'

lol

[night lake]

😮

i can finally use a bs above 10

[gray rover]

incorporating noobies' fixes n improvements too
so all in check

[gray rover]

Actually.. I haven't tested the limits butttt

from base 13 bs max, I could go up to 20

perhaps could do more. ( There's also 1 inplace operation more )

I suppose you could help the speed a lil with tensorfloat32 usage

or... improve the memory even more with memory fallback in nvidia ctrl panel

don't do it

it allows to fallback to extra normal ram memory instead of oom'ing
but you know the deal with ram memory

[night lake]

i wont
-# im on amd

[gray rover]

oh

ohhhhhh
That's new.. All this time I thought you're rocking on nvidia

damn

[night lake]

i wish 😭

[gray rover]

fffff, rip man

either way.. should help, regardless

[night lake]

yeah...

[glad nebula]

the new update also includes the inplace that reduces vram usage without checkpointing?

that alone allows 8 batch size on 8gb vram gpus in fp32 training

[gray rover]

yea, improved inplace is by default

checkpointing is optional from the ui level

https://github.com/codename0og/codename-rvc-fork-3/releases

3.0.1 up

bruhhh, can't promise how stable this is as it's just a test run but damn, even 23 works ( 9 bs higher than baseline oom (( bs 14 )) )
lmao

Either way. Cya around, going off

yup, works

[glad nebula]

huge update

[gray rover]

now, I can rest

peace

[glad nebula]

good night

[gray rover]

Gnight

[glad nebula]

[gray rover]

[night lake]

night king

[polar flax]

https://tenor.com/view/cat-clap-cat-jump-cat-kitten-mklpkl-gif-20733514

[elder willow]

hi

[broken spade]

hi any recco settings for nvidia rtx 2050

[worthy coyote]

annie are you okay

[shrewd surge]

how i can use rvc 2 models?

[elder willow]

https://youtu.be/oOBjntI2xK0?si=wPeezjzv63WqU_A5

[shrewd surge]

oh thanks

its better than weights ?

[upbeat quartz]

guys pls help. i used to make ai covers with google collab thingy like a year ago, now its dead. does someone have a new link to it? or how do you guys make your song covers

the link to old one i used: https://colab.research.google.com/drive/1Gj6UTf2gicndUW_tVheVhTXIIYpFTYc7?usp=sharing#scrollTo=7vh6vphDwO0b

[elder willow]

ive been trying to make a rvc model

with colab
but now with it being shut down

rip my rvc making

[shrewd surge]

.

[sleek lance]

hello, can someone please help me get the voice changer to work

ive been trying i cant seem to make it work

[hexed bay]

What seems to be the issue?

First of all did you install the version that is for your gpu?

[sleek lance]

the voice keeps cutting out for some reason, its like very delayed takes 10 seconds for it to say anything even though i said it

[sleek lance]

i downloaded for amd

i cant even speak to anyone with the voice changer

[hexed bay]

Is your GPU modern?

The chunk settings etc is what determines delay I think

#🔍│help-w-okada probably knows more

[sleek lance]

i dont know, maybe if you want to come on screenshare into a voice chat

[solar torrent]

If your PC has a GPU that's newer than or as GTX 10xx series, W-Okada should be able to run on GPU.

[sleek lance]

its not newer

[solar torrent]

Is your GPU lower than that?

[hexed bay]

I am not on my computer atm and probably won't be until tomorrow

[sleek lance]

let me check

[sleek lance]

its AMD Radeon(TM) Vega 8 Graphics @solar torrent

[solar torrent]

Oh wow, the AMD Radeon. This GPU was released in 2019, so it should be able to run fine. For AMD/Intel GPU, download the DirectML version of fork W-Okada.

[covert lake]

did you download wokada from a youtube video?

[solar torrent]

Oh wait, that's an integrated one. That's mean your PC doesn't have a dedicated GPU.

[sleek lance]

i downloaded it from huggingface.co

[sleek lance]

wait what

[covert lake]

did you see our written guides or did you use another tutorial

[covert lake]

yea it's bad

laptops are weak

[sleek lance]

its a pc* not a laptop

[sleek lance]

i used another tutorial

[covert lake]

that's bad then

youtube tutorials are old

[covert lake]

do you maybe see a GPU 0 or GPU 1?

[edgy bloom]

Congratulations Nick088 [ITA/ENG] by Weights!

Your Charizard is now level 72!

[covert lake]

else you didn't buy a dedicated gpu

[sleek lance]

gpu 0

[covert lake]

show a screenshot in #🔍│help-w-okada , maybe u got integrated and dedicated gpu

[sleek lance]

aight

[covert lake]

it sounds kinda weird u didn't buy a gpu for ur desktop

[north heron]

Sorry to butt in suddenly. But I would like to hear your stuff about the announcement of the o3 model.

I was expecting more action here but it looks like the people here are still getting up to speed on this, thats fine.

[solar torrent]

Nah, I'm too slow for a trend.

[weak cargo]

그림

[gray rover]

gpt just isn't as much of a big deal anymore ( or so it seems to me

[north heron]

You sure? Im seeing a haywire of reactions coming from X and other spaces.

I think we just need to wait a bit more here. I really want to hear what the folks here have to say.

[gray rover]

that's right, X

truth to be told, not a lot of people here are into X / Twitter I think but guess you right, we'll wait and see how it goes

[dusty glacier]

You mean 𝕏??

[covert lake]

I use X/Twitter 🔥

i seen it's pretty good but not like I can use it anyways lol

[gray rover]

oh, how come?

[river verge]

Wake up, everyone! A new RegalHyperus drum model just released!
Get Lucky (Drum model no. 556)

[covert lake]

u gotta fill a whole ass form for using o3 in early access

[analog cosmos]

o3 isn't publicly available yet, only way to get access is to apply for Early access safety testing
https://openai.com/index/early-access-for-safety-testing/

[covert lake]

ye

[gray rover]

dang

[north heron]

I know about the forms, IDK if my uni is aplicable as an org..

[north heron]

Doesnt need to be X in particular, even ops places like hacker news are applicable.

Just lookin for the wisdom of this crowd

[gray rover]

yuh, got it

[north heron]

Eksu

[river verge]

Wake up, everyone! A new RegalHyperus drum model just released!
Mor Ardain (Drum model no. 557)

[grim terrace]

Hii guys

It’s been a while since I participated in voice model making

Are there any new standards I should know about?

[gray rover]

Aside of recommending the use of forks with averaged loss
not much

[grim terrace]

Wish I knew what that means

[gray rover]

You used tensorboard before?

[grim terrace]

Yep

[crude night]

How to download?

[gray rover]

In that case, I'll simplify it to:
Tensorboard logs each epoch's loss based on ' each epoch last step's loss '

Average makes it so it logs additional metrics but averaged over custom N steps ( my approach )
or Noobies approach where it's averaged over some period / amount of steps? epochs? something like that

Visually looks like so

ex. My epoch ( each ) is 44 steps. Instead of logging a loss that depicts performance of epoch at the last step, it logs average ( in my scenario, every 11 steps so 1/4th of an epoch )

This should be... pretty straightforward ye

[grim terrace]

Uh huh

[gray rover]

In a short, helps to know how your training goes

[grim terrace]

Have the models from UVR improved at all?

[gray rover]

Naturally, mvsep too

mel-roformed based, bs-roformer based ( mvsep, not sure on uvr as I don't use it )
those are kings now

[grim terrace]

You have got to provide me with a dictionary

[gray rover]

Those are just names of models for separation ( and their architecture

[grim terrace]

oh what do you use

[gray rover]

personally? still bs-roformer is my pick ( even if some prefer mel one

You can test both really, mvsep's free

[vestal temple]

Hi

[gray rover]

https://mvsep.com/en/demo
Or you can see the results of users' separations in here

[grim terrace]

Ok!

[willow grotto]

Why when I do this the both models mix together and their faces look very weird?

[covert lake]

Because you're using 2 models at the same time lol

Maybe try to be specific on separating them and using each trigger word

[willow grotto]

Ok do you know a way I can separate the baking vocals from the song and make the ai model sing them?

[covert lake]

What's your PC GPU?

[willow grotto]

Why

I think it's Nvidia GTX 1660 Super

[covert lake]

Because you can run RVC (Retrieval-based-Voice-Conversion, Speech To Speech, the AI used for every AI covers also on Weights.gg) on your pc

[willow grotto]

I know that

I asked if there's a way I can separate the baking vocals from a song

[covert lake]

Your GPU is good enough to do inference (use models) locally (on ur pc), you won't be able to train (make models) but use them

You can:

• Locally (runs on your pc so the speed depends on that, you will have to set it up with the guides):
  • Applio: A fork of RVC with some extra features like Applio TTS, kinda faster and simpler but same quality tho
  • Mainline: The original RVC
• Cloud (remote good pc, easier and faster than ur PC but it's limited):
  • Ilaria RVC Zero: fastest and simplest that you can get for free
  • Weights.gg: Partnered with AI Hub, lets u do them easily but u may be in a queue
  • Applio Colab: max 4 hours, not granted, of GPU

Easiest possible (automatically separates vocals & instrumentals) : weights.gg
easiest cloud: Ilaria rvc zero
easiest local: Applio

[covert lake]

Weights.gg does that automatically when doing an AI Cover

else you would have to look at https://docs.ai-hub.wtf/rvc/resources/vocal-isolation/

[willow grotto]

Ok thanks

[covert lake]

Yw

[hardy mauve]

Hello, the virtual cable isn't detecting input voice or maybe it's some bug I don't know,

Can anyone help me with it please!

[covert lake]

Wrong channel, show me your settings in #🔍│help-w-okada , Frieza

[lean oyster]

rerew

[covert lake]

?

[wheat burrow]

chat

could I theoretically run rvc on a jailbroken iPhone

not the voice changer part just audio conversion

[gleaming hull]

Maybe. If you got the code for it.

[wheat burrow]

ok thanks

[gleaming hull]

That's not a definitive answer. But, you never know unless you try it. Since, I assume, there was no information on the internet.

[wheat burrow]

I could have just ran tailscale on my iPhone and pc but I’m kind of interested how this would run so I’m just gonna try it

[polar flax]

there should be termux lookalike for iOS, so you can run under it

[wheat burrow]

I was thinking about running the apple silicon version

I have bash and python installed and was going to look at the other dependencies after

[covert lake]

it is possible on Termux, but it's Android only

however, there's an unofficial iOS alternative to termux, like iSH and A-Shell

maybe try those, but the commands could be different

Also reminder that it's just a phone, and considering that iPhones except the last one, have only 6gb of RAM, it won't be the best

ofc running on a good pc or cloud will be better,

but this can be a funni experiment

goodluck

if u actually manage to do this, maybe it would be good u write a guide if the commands are different, so other iPhone users can do it too

[wheat burrow]

I have native shell running

no need for one of those apps hopefully