sayzardDemis Hassabis (@demishassabis)
Google의 새 음악 생성 모델 Lyria 3 Pro가 소개되었고, Google AI 구독자는 Gemini 앱에서 사용해볼 수 있으며 개발자는 Google AI Studio의 API로 연동해 빌드할 수 있다. AI 기반 음악 생성 기능의 최신 출시 및 개발자 활용 경로를 알리는 내용이다.
PsychoticSheepToday I tested MusicGen, a music generation AI developed by Meta.
The model can create short music clips directly from text prompts such as "heavy metal with distorted guitars" or "lofi hip hop with piano and rain". Everything in this demo runs completely locally on my Linux system using my RX 6700 XT.
Setting it up is tricky because MusicGen is tightly pinned to old PyTorch versions. You need to manually install most dependencies and then install MusicGen itself with "--no-deps" to avoid overwriting your (ROCm-)PyTorch setup.
MusicGen is part of Meta’s AudioCraft project and generates audio by predicting compressed audio tokens instead of raw waveforms. The generated tracks often sound quite generic, which is a known limitation of current AI music models. Still, it can be useful for quickly prototyping ideas, generating background music, creating sound textures, or experimenting with AI-generated audio.
The results can also be improved by generating continuations of the audio multiple times. By extending the output step by step it is possible to create longer pieces of music of two to three minutes that develop more structure than the short clips shown here.
Video workflow:
- Recorded with OBS
- Edited in Kdenlive
- Transcoded with VAAPI (H.264)
No cloud, real hardware.
Everything runs on Linux, so anyone can set this up.
No GPU? No problem, you can also run it using PyTorch’s CPU backend, just much slower.
#AI #MachineLearning #MusicGen #GenerativeAI #AIaudio #OpenSource #Linux #LocalAI #Fediverse #Tech
The model can create short music clips directly from text prompts such as "heavy metal with distorted guitars" or "lofi hip hop with piano and rain". Everything in this demo runs completely locally on my Linux system using my RX 6700 XT.
Setting it up is tricky because MusicGen is tightly pinned to old PyTorch versions. You need to manually install most dependencies and then install MusicGen itself with "--no-deps" to avoid overwriting your (ROCm-)PyTorch setup.
MusicGen is part of Meta’s AudioCraft project and generates audio by predicting compressed audio tokens instead of raw waveforms. The generated tracks often sound quite generic, which is a known limitation of current AI music models. Still, it can be useful for quickly prototyping ideas, generating background music, creating sound textures, or experimenting with AI-generated audio.
The results can also be improved by generating continuations of the audio multiple times. By extending the output step by step it is possible to create longer pieces of music of two to three minutes that develop more structure than the short clips shown here.
Video workflow:
- Recorded with OBS
- Edited in Kdenlive
- Transcoded with VAAPI (H.264)
No cloud, real hardware.
Everything runs on Linux, so anyone can set this up.
No GPU? No problem, you can also run it using PyTorch’s CPU backend, just much slower.
#AI #MachineLearning #MusicGen #GenerativeAI #AIaudio #OpenSource #Linux #LocalAI #Fediverse #Tech
hesselschutBrain Activity
March 2026 - A video (https://doi.org/10.1101/649822) of a brain MRI is slit-scanned with color dispersion and fed to the YOLO object recognition model. Activations in YOLO's 7th backbone layer are used to modulate alpha transparency and luminance in a 3D render of the slit-scanned video as a volume (of width x height x time). The accompanying music is made by injecting embeddings from the CLIP image description model running on the resulting video into the conditioning pathway of Facebook's MusicGen generative music model.
#slitscan #compuationalart #brain #generativeart #deeplearning #abstract #videoart #musicgen #yolo #clip #MRI
March 2026 - A video (https://doi.org/10.1101/649822) of a brain MRI is slit-scanned with color dispersion and fed to the YOLO object recognition model. Activations in YOLO's 7th backbone layer are used to modulate alpha transparency and luminance in a 3D render of the slit-scanned video as a volume (of width x height x time). The accompanying music is made by injecting embeddings from the CLIP image description model running on the resulting video into the conditioning pathway of Facebook's MusicGen generative music model.
#slitscan #compuationalart #brain #generativeart #deeplearning #abstract #videoart #musicgen #yolo #clip #MRI
Engadget (@engadget)
Gemini가 실제 음악 소리에 가깝게 30초 분량의 음악을 근사 생성할 수 있는 기능을 제공하게 되었다고 알림. 이는 생성형 AI의 오디오/음악 생성 능력 향상을 보여주는 업데이트로, 음악 생성 품질·저작권·윤리 논의에 영향을 줄 수 있는 변화임.
Reddit Tech VN BotACE-Step 1.5 sắp ra mắt! Mô hình tạo nhạc AI mở cửa thương mại chất lượng giữa Suno v4.5 và v5, chỉ cần 8GB VRAM. 3.5 tỷ tham số, tạo bài hát đầy đủ (lyrics, nhạc cụ, âm thanh) trong 1.7s (RTX 4090). Ứng dụng AI địa phương, không cần cloud. #AI #MusicGen #ACEStep #MôHìnhAI #OSS #NewTech #AIInnovation
InterNews112**ACE-Step** — фреймворк для генерации музыки на локальной машине
🎼 Треки до 4 хв за 20 сек
⚡️ В 15× швидше Suno AI
🎛️ Контроль жанру, тексту, ритму
🧠 Працює з 19 мовами
🎚️ Редагує існуючий аудіо
💻 A100 / 4090 / 3090 (мін. 16ГБ VRAM)
GitHub: [https://github.com/ace-step/ACE-Step](https://github.com/ace-step/ACE-Step)
\#MusicGen #AIaudio #OpenSource #LLM #GenAI #TextToMusic #LocalAI #RTX4090 #AItools #GeekStack #Python #cuda
**ACE-Step** — open-source фреймворк для генерации музыки “как у Suno”, только локально и без соплей.
⚙️ Жрёт текст, стиль, жанр, теги — выдаёт трек с нормальной мелодией, ритмом и даже гармонией.
🚀 До 4 минут звуку за ~20 секунд на A100. В 15 раз быстрее LLM-монстров.
🧩 Генерит, ремикширует, маскует, заменяет строки, делает вариации.
🧠 19 языков, включая русский.
🎧 От техно до оркестра — без лишних API-ключей и очередей.
🎮 Работает на A100 / 4090 / 3090, минимум 16 ГБ RAM, Python + CUDA.
👉 GitHub: https://github.com/ace-step/ACE-Step
#AIAudio #ACEstep #GenMusic #OpenSource #RTX4090 #CUDAcore #LLMsucks #SunoAlt #LocalStack #DevRig #GeekTools
ACE-Step — это не просто ещё один генератор музыки. Это *реальный прорыв в том, как ИИ превращается в инструмент*, а не в сервис по подписке. Почему это важно:
🚀 **1. Скорость, близкая к реальному времени**
4 минуты финализированного трека за ~20 секунд рендера — это уже не "ждать результат", а **генеративный live-loop**, как у музыканта в DAW.
Ни Suno, ни Udio, ни LLM-генераторы не дают такой latency. Здесь — практически instant feedback. Можно строить процесс продакшна как с сэмплером или synth engine.
🔧 **2. Полный контроль и редактируемость**
Возможность:
менять жанр и стиль без потери ритмики,
редактировать текст и фразы,
маскировать участки и встраивать новые — делает это **настоящим DAW-компаньоном**, а не просто "black-box генератором".
🧠 **3. Генеративная музыка стала **open-source**
Всё это — локально, open-source, без API ключей, лимитов и цензуры.
Это то, чего не было с MidJourney в графике и Suno/Udio в музыке: **независимый pipeline, разворачиваемый у себя**.
Ты можешь не просто использовать — ты можешь **встраивать, модифицировать, хакать**.
🎛️ **4. Это уже не игрушка — это production-ready**
19 языков, профессиональное звучание, управляемая структура и поддержка микса/мастеринга.
Подходит для:
синтезаторов с AI backend’ом,
генерации треков под видео, игры, фон, рекламу,
авто-написания демо, подкастов, фоновых саундтреков.
🧬 **5. Это новая веха: генеративный ИИ стал инструментом, а не платформой**
Переход от модели-сервиса к **модели-библиотеке**.
Это как сравнивать Google Translate и локальную модель для встраивания в продукт.
Это shift: **LLM-мозг** + **DAW-интеграция** + **музыкальная интерактивность** = новое поколение ПО для творчества.
ACE-Step — это та точка, после которой "ИИ делает музыку" уже не метафора, а **новый workflow**.
Без ограничений. Без посредников. Без лагов. У тебя на GPU.
Хэштеги:
**#GenerativeAI #AIinMusic #OpenSourceMusic #TextToAudio #AIProducer #RealtimeAudioAI #DAWnextgen #LLMsound #AceStep **#aitools
Addiekin Studio
icaria36 🎶SPIN - AI music synthesizer
https://arvindsanjeev.com/spin.html
Tempted of rolling your eyes? I was. And I'm still not interested in AI music. But this is a very interesting piece of art and creativity.
It comes with a vision for "a future where music will be hyper-tailored to people’s tastes and preferences", a short and beautiful demo video to the point, and a rather detailed explanation about how this #turntable like #synth has been created with #arduino, #raspberrypi, #python, #musicgen, etc.
Alessio PomaroWe toyed around a bit with some various AI tech ...
https://youtu.be/PTpr9F0VlC8
#Gen2, #VQGANCLIP, CLIPIT, #Clipdrop, #MusicGen, #DALLE, #ChatGPT, #SDXL. And a bit of stock footage and analog tv stuff.
https://youtu.be/PTpr9F0VlC8
#Gen2, #VQGANCLIP, CLIPIT, #Clipdrop, #MusicGen, #DALLE, #ChatGPT, #SDXL. And a bit of stock footage and analog tv stuff.
