The Music-Generative Usable+ AI (MusGU+) framework is a musician-centered evaluation framework designed to assess how generative music models can be adapted, used, and controlled in real-world creative contexts. The framework evaluates models along three complementary dimensions, with each dimension addressing a key question from the musician's perspective:
📖 Read the detailed evaluation criteria, return to the discovery tool or inspect the model's YAML source file.
audio synthesis, style transfer
Custom model training does not require local compute resources. Morpho’s training pipeline is fully cloud-based, requiring no GPU, no local installation, and no ML environment, which removes hardware barriers entirely for end users.
Morpho is designed to be trained on small, musician-scale datasets, with clear guidance suggesting ~45 minutes of unique audio as a practical target. This scale is well within reach of individual musicians working with personal recordings or sound libraries.
A complete and practical adaptation pathway is provided through Neutone’s proprietary training service, allowing musicians to upload audio and train custom models end-to-end. However, the training process is closed-source and platform-bound, with no access to underlying training code, checkpoints, or local fine-tuning workflows.
Model adaptation is designed explicitly for non-technical users. Training is performed via a drag-and-drop, no-code interface, with extensive musician-oriented guidance on dataset preparation, making the process accessible to users without programming or machine learning experience.
Trained models can be freely used by their creators within the Neutone ecosystem, but redistribution is constrained to the platform. Models cannot be exported or reused outside Neutone’s plugin environment, limiting portability despite creative freedom within the system.
Neutone Morpho is distributed as a polished VST3/AU plugin with a dedicated graphical interface, designed for immediate use inside major DAWs. No scripting, external inference setup, or technical configuration is required.
The plugin itself is freely downloadable with a limited set of bundled models and a time-limited trial of the full library. Access to additional pretrained models and custom model training requires account creation and paid purchases, introducing moderate platform-level restrictions.
Morpho is explicitly designed for real-time audio processing, operating with low and predictable latency suitable for live performance. The system is optimized for consumer CPUs (e.g., Apple M1-class hardware) and is demonstrated in both DAW and embedded hardware contexts.
Morpho integrates directly into standard music production workflows as an audio effect and instrument plugin, with automation-ready parameters and compatibility with all major DAWs. Ongoing work (e.g., Project LYDIA with Roland) further demonstrates its suitability for hardware and live performance contexts.
Audio output generated using Neutone Morpho is fully usable for personal and commercial purposes. Models are trained exclusively on licensed or user-provided data, and users retain ownership of sounds they create.
Neutone maintains an active Discord community, direct support channels, detailed blog posts, FAQs, and ongoing artist-focused communication, providing accessible support for both creative and technical questions.
Morpho conditions on incoming audio, using learned representations to transform signals in real time. While rich in effect, conditioning is limited to a single musically meaningful modality rather than multiple symbolic or structured inputs.
Morpho provides continuous time-varying control through real-time audio resynthesis. The output dynamically follows the overall temporal shape and performance characteristics of the incoming signal while transforming it through the selected timbral model.
Neutone Morpho is designed for timbre and style transformation, using the incoming audio to drive generation through a model trained on a different sound domain. However, disentanglement is not explicitly enforced in the learned representations, and latent variables are not guaranteed to correspond to independently interpretable musical attributes.
Neutone Morpho exposes both model-specific macro controls and an experimental Micro View where users can directly offset and scale latent variables. Additional preprocessing and postprocessing controls, including pitch shift, input filtering, dry/wet mix, and output filtering, provide further real-time control over the model's behavior and output.