Third-party software, on-device AI, and services

Everything Light Sound Colors uses that we did not write ourselves, what it does, and where it runs. Last reviewed 20 August 2026.

Third-party code we redistribute

The shipped app includes one vendored third-party runtime family: Google MediaPipe Tasks Vision and its required model and WebAssembly components. It uses no runtime CDN, analytics library, remote font service, or cloud AI service; MediaPipe is served from this site.

Apache 2.0 · BSD-3-Clause · MPL-2.0

Google MediaPipe Tasks (Vision)

© Google LLC. Used for on-device hand, face/iris, and body-pose landmark estimation. We ship the runtime and the model files with the app so nothing is fetched from Google at play time.

The complete licence texts, component notices, and corresponding-source locations ship with the app at /mediapipe/license, /mediapipe/notice.txt, /mediapipe/third-party-licenses.txt, and /mediapipe/source-availability.txt.

CC0 1.0 · Public Domain

Instrument recordings

The Pad Machine's drum sounds and the sampled instrument voices (grand piano, marimba, xylophone, glockenspiel, tubular bells, recorders, kalimba, vibraphone, harpsichord, tenor saxophone, concert harp, Dan Tranh zither, FM keys) are real recordings dedicated to the public domain (CC0 1.0) by their makers. We ship them with the app; nothing is fetched from a third party at play time. Sounds you record with the Sampler stay on your device (see the Storage Notice).

Per-file origins, licence evidence, and SHA-256 checksums ship with the app at /samples/SOURCES.md. CC0 requires no attribution; we credit the makers because the sounds are good.

Artificial intelligence in this app

Light Sound Colors uses machine-learning models. We think you should be told that plainly rather than have to infer it.

What the models do

The four shipped MediaPipe model files (three model types) are neural networks. They take a camera frame and estimate the coordinates of points on a hand, a face, or a body. Those coordinates move a marker and trigger a note. That is the whole job.

Where they run

Entirely in your browser, on your own device. Frames are processed moment to moment and discarded. Nothing is uploaded, and no external AI service is contacted at any point during play.

What they do not do

Camera input is always optional. Touch play uses no models at all.

Landmark estimation is not perfect and is not meant to be. Lighting, clothing, distance, and disability all affect how well a model tracks. If it tracks you poorly that is a limitation of the model, not a statement about you, and every camera feature has a touch or switch alternative.

Services we rely on

These are hosted services, not code bundled into the app. They are listed here because they process data on our behalf; see the Privacy Policy for the detail and your rights.

Processor

Netlify

Hosts the site and runs the small server functions that verify a Pro licence. Standard web-server logs apply.

Processor

Gumroad

Handles the one-time Pro purchase and is the record of that transaction. We never see or store your card details. If native app-store versions are released, the store's own billing will be used there instead.

What we do not use

© 2026 Halal Helper LLC d/b/a Omerica Studios. All rights reserved. Light Sound Colors™ · Patent pending · Not a medical device · Use at your own risk.