Moisés Horta Valenzuela of Berlin-based SEMILLA.AI Studio wrote in to announce the release of a neural synthesizer that lets users cast seeds into a latent sound field and listen to what grows.
SEMILLA.AI v2 is a divination instrument for sound. Audio on your track is encoded into the latent field of a neural audio model and re-grown, in real time, as a new voice — two models at once, one per channel. You steer the timbre by casting seeds: gravity wells that pull a swarm of green particles whose positions in the field are the sound.
SEMILLA.AI is not another preset synthesizer — it is an oracle you consult. Hand the reading to a physics simulation and chaotic orbits — Circle, Lorenz, Chua, Halvorsen — and the constellation composes melodic and textured patterns that never exactly repeat. Freeze States, morph between them with an LFO, and sculpt the return with tape Delay, shimmer Reverb and a latency-compensated Dry/Wet.
SEMILLA.AI v2 features
- Real-time neural synthesis: Incoming audio is encoded into a neural audio model’s latent space and decoded back as new sound: live timbre transfer inside your DAW.
- Two voices at once: Independent neural audio models on the left and right channel; Browse loads your own trained models.
- Seeds & particles: Seeds are gravity wells (Pull / Pole / Spin); green particles carry the latent dimensions, and their positions are what you hear.
- Physics as performance: Gravity, Friction, Bounce, Turbulence, Speed and Range set how the swarm wanders.
- Compose through orbits: Circle and strange attractors (Lorenz, Chua, Halvorsen…) tow the seeds; a per-seed X/Y Pattern matrix writes melodic and textured patterns hands-free.
- Latent manifolds: Selectable mapping modes reshape how seed motion traverses the latent space, from gentle interpolation to exploratory jumps.
- Per-dimension control: Radial Bias and Scale controls for every latent dimension, with randomize, reset and flip per side.
- States: Capture seed constellations, mark A/B, and morph between them with a shaped LFO (rate and depth).
- FX chain: Stereo width, tape Delay (per-side time, tempo sync, ping-pong), shimmer Reverb (Size / Decay / Damp / Diffusion) and a final latency-compensated Dry/Wet.
- Runs headless: Orbit and modulation keep moving with the window closed.
- Presets: Save and recall the whole reading: models, constellation, physics, FX.
SEMILLA.AI v2 for Windows and macOS (VST3) is available to purchase for €49 EUR (Complete license with 7 AI models) or €99 EUR (Extra License with 12 AI models). A 7-day trial version is available to try the synth for free.
More information: SEMILLA.AI Studio