About Nila and shipped components

Nila is a fictional, non-commercial research demo. She listens and reacts without speaking. The photoreal head was made with non-commercial research tools and is shown here for non-commercial research only.

The public demo ships these components and assets:

• MHR mannequin body and hands — Meta Momentum Human Rig, Apache-2.0.

• FLAME 2023 Open face parameters — CC-BY 4.0. FLAME credit: Li, Bolkart, Black, Li, and Romero, “Learning a Model of Facial Shape and Expression from 4D Scans,” ACM Transactions on Graphics 36(6), 2017. The browser mannequin uses the CC-BY FLAME 2023 Open model.

• MediaPipe Tasks Vision 0.10.32 and the Face Landmarker v1 model — Apache-2.0. The face model runs locally in the browser.

• MediaPipe Tasks Audio 0.10.32 and YAMNet v1 — Apache-2.0. YAMNet is trained on AudioSet and runs locally in the browser.

• three.js 0.186.1 — MIT.

• On-device Whisper — the speech model and Transformers.js runtime are excluded from this public build. The public speech choices are cloud speech recognition, optional Chrome speech, and text input.

• Room background photograph — Photo by SHHOONYA The Design Studio on Unsplash, under the Unsplash Licence.

• Photoreal Gaussian head — made from non-commercial research assets and methods. It is included only for this non-commercial research demo.

For the project-wide release plan and current licence decisions, see [docs/RELEASE.md](../docs/RELEASE.md).