UNDER THE HOOD

Text-To-VST isn't just an interface. It's a patent-pending system combining a local neural network with a custom-engineered DSP audio engine.

MODULE 01

Spatial Reverb Engine

A lush, algorithmic reverb capable of generating everything from tight drum rooms to infinite frozen voids.

  • Stereo Width Hack: Proprietary algorithm for ultra-wide imaging.
  • P4 Freeze Mode: Infinite decay capability for ambient drones.
  • Damping Control: Maps "Dark" vs "Bright" prompts to HF absorption.
MODULE 02

Shaping Distortion

More than just clipping. This module uses a pre-emphasis filter chain to emulate the behavior of analog overdrive pedals.

  • "Tube Screamer" Topology: High-pass pre-filter + Low-pass post-filter.
  • P4 Bit-Crush: Variable bit-depth reduction for Lo-Fi textures.
  • Tanh Saturation: Smooth, analog-style soft clipping.
MODULE 03

Adaptive Dynamics

A workhorse compressor designed for speed. Features intelligent gain staging so your volume doesn't drop when you squash the signal.

  • Auto-Makeup Gain: Automatically compensates volume based on Threshold.
  • Parallel Mix: Built-in "New York Compression" blend knob.
  • Extended Release: P4 controls release times from snappy (10ms) to glued (500ms).
MODULE 04

3-Band Tone Sculptor

A musical equalizer tuned with gentle Q-curves to mimic passive hardware EQs.

  • Baxandall Curves: Wide Q factors (0.5) for smooth Low/High shelving.
  • P4 Air Band: A dedicated 12kHz shelf boost for instant clarity.
  • Musical Mid: Proportional Q peak filter for cutting mud or boosting presence.

The AI Architecture

Feed-Forward Network

Our custom PyTorch model is optimized for CPU efficiency. It uses a lightweight architecture to map semantic tokens directly to float parameters.

ONNX Runtime

The industry standard for high-performance inference. By embedding the model as binary data, we ensure 100% offline functionality.

Hybrid Dataset

Trained on 50,000+ verified presets. We combine synthetic generation with human verification (RLHF) to ensure professional results.