ꜱᴇᴍɪʟʟᴀ ᴀ.ɪ.

Today we’re releasing SEMILLA.AI v2, a totally new interface reconceptualization from the original SEMILLA.AI v1, released in 2023.

🔊 We are the first neural audio synthesis plugin to transform the latent space into an instrument. 🌱 The interface is designed for experimentation and exploration of the neural networks black box; no text prompts or other conditionings, just chaotic system dynamics.

neural network music > ai music 💚💚💚

SEMILLA V2 (@semilla_ai) by @hexorcismos turns the latent space of RAVE audio models into a VST3 performance surface. Seeds pull and repel particles; their motion changes the resynthesized sound in real time. sonicfield.org/semilla-v2-neu…

Further than becoming a timbre-transfer tool (which is still available), SEMILLA.AI v2 is made to perform the AI without any audio input, only through the latent trajectories generated by the chaotic systems simulation. I’m very happy to see what you make with it!

Seeds now act as attractors or repellents within a physics simulation, where a mesh of particles are affected by gravity, turbulence, friction, bounce and range. The trajectories from the mesh particles within this simulation are then fed to a decoder-only neural audio synth.

Today we’re releasing SEMILLA.AI v2, a totally new interface reconceptualization from the original SEMILLA.AI v1, released in 2023.

We have some news for you this week, here’s a sneak peek…

We are really excited about the potential new sonic expressions that these tools will enable for the live context, installation, performances and other new spaces ! What will be the new music styles? Let’s hack these tools & listen! 🔊🔊🔊

The tools are open: ❉ train on your own music → CoDiCodec-Flow ❉ jam today → Magenta RealTime 2 apps ❉ the accompanist → LiveBand ❉ stream your diffusion models → LMDM ❉ patch into Max/Pd → RAVE

Lately, we been been feeling a shift in AI music. The latest most popular systems are offline: you write a prompt, you wait, a finished, derivative .wav is generated. Now the models are learning to play in time with us. En vivo. Here’s survey of the new realtime wave 🧵👇

Just added a Google Colab notebook to the CoDiCodec-Flow repo. You can train your own real-time neural audio synthesis model with about 2-3 hrs. of audio and get results in about 3 hours of training time. Small Data models > Corporate Big Tech models github.com/moiseshorta/Co…