Generate a plurality of sonic constellations with neural audio synthesis ⚫️ Developed by @hexorcismos 🌱 Contact at [email protected]
Berlin, Germany
Joined July 2022
- Tweets124
- Following1
- Followers718
- Likes111
Pinned Tweet
🌫️ Introducing NEBULA v1, a latent space audio sampler and concatenative synth.
Drop in your sounds and NEBULA maps them into the latent space of a neural codec, arranging them by timbre so you can drift through, select and stitch fragments into new textures.
ꜱᴇᴍɪʟʟᴀ ᴀ.ɪ. retweeted
'neural-play' is a Pure Data playground for learning about artificial neural networks.
you can play live interactive sound examples of the McCulloch–Pitts neuron all the up to VAE, GANS, transformers and diffusion, all in vanilla Pd
github.com/moiseshorta/neura…
🌱🌫️ Friendly reminder to Windows users 🔊
🌫️ NEBULA v1 is now out on Windows !
We’ve been working hard these past weeks to complete NEBULA’s Windows release, importing the hand-tracking and optimizing the neural audio codecs; and we’re finally ready!
🔗 Download NEBULA now!
semilla.ai/software/nebula
ꜱᴇᴍɪʟʟᴀ ᴀ.ɪ. retweeted
🌱🔊 See you tomorrow (21 of Oct. 2026) at #SilentGreen Berlin.
I’ll be performing a new live set in quadraphonic multi-channel, sharing the night with @ConcepcionHue13 .
Come through! DM me if you want friend discount price
🌫️ NEBULA v1 is now out on Windows !
We’ve been working hard these past weeks to complete NEBULA’s Windows release, importing the hand-tracking and optimizing the neural audio codecs; and we’re finally ready!
🔗 Download NEBULA now!
semilla.ai/software/nebula
🧪 Introducing our first SEMILLA.AI Studio Research proceeding:
🤖 🔊 Agentic Neural Audio Synthesis:
We gave an LLM agent a frozen neural audio codec and one request: make it speak. No encoder, no data, no TTS, and it can't hear.
Our preliminary results... 🧵
Global optimisation couldn't reach the consonants, so the agent invented local latent surgery: re-search the 2 frames a short window decodes exactly, splice the winner.
Over 23 logged iterations: harmonics-to-noise −6.5 → 17.9 dB, vowels identified 10% → 90–100%. ...
🧪 You can find the full write up report, paper and code on our website:
semilla.ai/research/agentic-…
🌫️ NEBULA v1 Feature Spotlight:
👂🏽LISTEN mode
Did you know that with NEBULA you can reconstruct a live audio signal input with your audio corpus?
You can alter the temperature, which gives the reconstruction more options and variety on which sample to reconstruct from.
🔗
🌫️ Download NEBULA v1 now, pop in the code NEBULA15 for an additional 15% off the launch price (until November 7)
semilla.ai/software/nebula
🔊 @RichardDevine doing some crazy work with NEBULA v1, proper latent space concatenative goodness !
🌫️ NEBULA grew out of our 2025 research with @tamlablinz (Tangible Music Lab, Linz) into neural synthesis and latent space control, first presented at SMC 2025 as “NEBULA: A PCA-Based Method to Explore RAVE-Encoded Audio Representations.”
🔗 Download · semilla.ai/software/nebula
🌱 We are bringing back the classic SEMILLA.AI direct latent space control.
🔊 We’ll be dropping this new feature, and others, as part of SEMILLA.AI v2.2 this upcoming week!
🚨Last day to get 35% off any license !
Use SEMILLASUMMER at checkout via semilla.ai/software
🌱 Stay tuned for our next audio plugin release in October!
🌱 Two more days for redeeming the SEMILLASUMMER discount code.
Get 35% off on any license tier until September 22 !
Download at semilla.ai/software
