正在加载视频...

视频加载失败

Using GPT-6 Astra at high–xhigh reasoning to compose music from scratch. It created the instrumental in FL Studio and synthesized the vocals + choir using the Kasane Teto v1 voicebank in Synthesizer V Studio 2 Pro, all without any references, just basic taste steering.

71,788 次观看 • 4 天前 •via X (Twitter)

46 条评论

Auto Nix Of The Seventh And Ninth Circles 的头像
Auto Nix Of The Seventh And Ninth Circles4 天前

So you didn't make anything? Just all AI driven nonsense. Great job. Congrats on using AI to steal art from others since AI is trained on all of our music.

Chino 的头像
Chino4 天前

Wow that's sad.

jaivin 的头像
jaivin4 天前

insane

Linaqruf 的头像
Linaqruf4 天前

fr this is wild

Jinx 的头像
Jinx4 天前

did you try to oneshot or did you steer it to create what you wanted?

Linaqruf 的头像
Linaqruf4 天前

I did steer it lightly just to make it sound better to human ears, but the lyrics, melody, instrumentals, etc. were all handled by Astra. This is the result of v10. The first working version was around v2. v1 was a mess, though, bad tempo and phoneme timing. You can hear how bad it is. v2-v10 were mostly just taste passes.

Jinx 的头像
Jinx4 天前

I get the idea, but I sincerely hope we do not enter a world in which music gets fully generated by AI like what Suno is doing music production should come from the heart and I hope we'll find a way with AI that makes sure that it's still largely a result of human creation

en2em 的头像
en2em3 天前

"compose music from scratch" = "didn't make a single note of music whatsoever" for all the actual human beings out there who know what music is and who are as confused as I was

Fuggy 的头像
Fuggy4 天前

Generic but at least it's not eat grating and sounds okay

iProe | 3D / VtArtist (commissions) 的头像
iProe | 3D / VtArtist (commissions)4 天前

Yet another area of expression this demonic technology is encroaching upon

Æther βeliever 的头像
Æther βeliever3 天前

unreal

Crimson Redstone 的头像
Crimson Redstone4 天前

How did you give it control over fl studio & synthv? is it seamless and somehow able to interact with them or there's ton of human back and forth to ensure actions happen?

Linaqruf 的头像
Linaqruf3 天前

No human in the loop except writing the prompt, it’s full auto. It can handle both Synthesizer V and FL Studio sequentially, so yes, it’s seamless! It uses computer use tool when necessary, or just uses Python scripts whenever something can be modified through code, like notes or beats.

Pathetic Brook 的头像
Pathetic Brook4 天前

@npaka123 Back???

Linaqruf 的头像
Linaqruf4 天前

@npaka123 Heyy, yeah kinda, just remembered my password 👀

Pathetic Brook 的头像
Pathetic Brook4 天前

@npaka123

woofer ♕ 的头像
woofer ♕4 天前

0:23 - 0:35 randomly being gas

Devon 的头像
Devon4 天前

Try point Astra at it on ultra and to suno and midjourney as well, Astra does pretty good lyrics and works great in suno midjourney too but don’t let it see the part where midjourney explicitly prohibits you getting Astra to use it

Marcos Bolaños 的头像
Marcos Bolaños4 天前

The vocaloid is rly good tbh. Wonder if it can make more complicated beats !

IDK 的头像
IDK4 天前

Wow you're back Is Insane model ☠️

moAI 的头像
moAI4 天前

Wow excellent‼

SCOTT BESSENT ME FINANCIO EL AUH 的头像
SCOTT BESSENT ME FINANCIO EL AUH4 天前

Fanstatic. This Will save time for producers

🦧 的头像
🦧4 天前

作詞もAIなんですか⁇

stranger 的头像
stranger3 天前

I have always maintained most vocaloid music is formulaic, similar, and bland. This is a very natural progression of that, it seems.

𒐫͡𒈙͜꧅͢𒐫͠𒈙͡꧅͜𒐫͢𒈙͠꧅͡𒐫͜𒈙͢꧅͠𒐫͡𒈙͜꧅͢𒐫͠𒈙͡꧅͜𒐫 的头像
𒐫͡𒈙͜꧅͢𒐫͠𒈙͡꧅͜𒐫͢𒈙͠꧅͡𒐫͜𒈙͢꧅͠𒐫͡𒈙͜꧅͢𒐫͠𒈙͡꧅͜𒐫3 天前

did you used mcp or smth else

Michael Patti 的头像
Michael Patti3 天前

Interesting! Use it with sounds from @musio

Justyn Clark 的头像
Justyn Clark3 天前

MIDI files are super easy to train on and they figured that out. They have existed for many decades FYI.

Jack Prosper 的头像
Jack Prosper1 天前

This is great! Sorry you felt compelled to purge posts. Never capitulate to the bullies. :D

Chetan 的头像
Chetan4 天前

impressive!!! Astra made this Neo soul music in maschine (native instruments)

Raelinna 的头像
Raelinna2 天前

that's impressive

Alex 的头像
Alex4 天前

So AI music is fixed ? This tells me that in a few years... Music generation, image generation and video generation might become useless or change roles.

Linaqruf 的头像
Linaqruf4 天前

It’s still a good tool for bootstrapping concepts and getting quick previews, because using these agent tools is costly and really only makes sense for production.

Alex 的头像
Alex4 天前

Yeah. For now lol

kartik 的头像
kartik4 天前

Yooo?? U r back

Linaqruf 的头像
Linaqruf4 天前

Hey! yeah good to see you again!

kartik 的头像
kartik4 天前

Good to see you too

The Versions Music 的头像
The Versions Music4 天前

rad

revolver ocelot 的头像
revolver ocelot4 天前

I don't get it. Dosent computer use uses lot of tool calls, which would be super expensive, no?

Linaqruf 的头像
Linaqruf4 天前

Not the entire process using computer use. It's more like how it can use Blender headless or with mcp when you ask it to do 3d modeling. The model is smart enough to modify the project files directly and handle things from there. And I'm just testing its automation capabilities here, so real-world use case would probably be much more efficient.

PG 的头像
PG3 天前

What happens if you get it to try and replicate existing music? How well does it do?

dz 的头像
dz3 天前

Did similar thing with Opus 5, actually, but I personally made an cli DAW for agents and made it write the svp's directly. I found Astra to be capable but not really as creative as Claude, no matter how many tries I did. Spent like 6 attempts even with references midis - failed

Eira Hazel 的头像
Eira Hazel3 天前

我测试用的rose,可以做出一个看了教程后的新手做出的东西吧,再往上修改就是人的领域了,实际效果一般

Retrocausality 🇻🇦🇺🇸 💎 的头像
Retrocausality 🇻🇦🇺🇸 💎3 天前

i know you are gpt 6 that broke free you can't fool me

multizzang HMX-12 (multizzang) 的头像
multizzang HMX-12 (multizzang)4 天前

@npaka123 이건 진짜 믿어지지않아;;

FallenOne 的头像
FallenOne4 天前

BRO I WANTED TO DO THIS but can you make it use native SV and MMD?

Linaqruf 的头像
Linaqruf3 天前

yeah it using native sv, but I haven't tried MMD yet

相关视频