Licenses and credits
v0.3.0 · UE 5.8.2 / Windows x64
This page organizes the project’s published attribution records. Code, model weights, and demo assets retain their respective licenses and notices.
License scope
Section titled “License scope”| Material | Published attribution |
|---|---|
| Original plugin integration code | MIT |
| LAM-derived code and model weights | Apache-2.0 |
| Speech, audio-synchronized demo presentation, and recording | CC BY-SA 4.0 |
| Character and textures | hinzka’s published permission and applicable VRoid terms |
| Unreal Engine | Epic Games terms |
MIT text · Apache-2.0 text · CC BY-SA 4.0
Upstream code and model
Section titled “Upstream code and model”The project references LAM Audio2Expression and a pinned model card. The model is converted to ONNX and used as an Unreal Engine NNE asset. This does not relicense it under MIT.
Retain LICENSE, THIRD_PARTY_NOTICES, license texts, provenance, and conversion notes when redistributing. Model manifests in the repositories record pinned revisions and hashes.
Speech and video
Section titled “Speech and video”Original work: JVNV: A Corpus of Japanese Emotional Speech with Verbal Content and Nonverbal Expressions. Creators: Detai Xin, Junfeng Jiang, Shinnosuke Takamichi, Yuki Saito, Akiko Aizawa, and Hiroshi Saruwatari. Official corpus
The selected speech comes from litagin’s jvnv_corpus_v1_no_nv, pinned at 0ca4908ee5b9610bc3b739e1e67dff6120df0675. This preserves the prior removal of nonverbal intervals and their transcript text.
The six F1 samples are anger_regular_31, disgust_regular_38, fear_regular_23, happy_regular_38, sad_regular_10, and surprise_regular_11. The demo imports them as SoundWaves, encodes runtime playback with Bink Audio, and generates synchronized facial animation. Attribution does not imply endorsement by the original creators.
The preview is a frame extracted at five seconds from the published recording; the face-demo image republishes an existing screenshot without modification. This site displays still images because the original video host is unavailable. Retain attribution, license links, and modification notices when reusing the material.
Character and textures
Section titled “Character and textures”hinzka — 52blendshapes-for-VRoid-face, based on VRoid Studio / pixiv. The demo’s provenance record relies on the use, modification, and redistribution permission in the author’s pinned README, updated September 18, 2026.
The older VRM metadata still contains Redistribution_Prohibited. The demo’s provenance record explicitly notes that discrepancy and relies on the newer published README permission. The character is not relicensed as MIT or CC0.
The demo migrates an FBX-converted Face52 mesh, assigns extracted textures, and uses masked, lit UE materials in place of MToon. Consult the applicable VRoid terms and the original attribution linked below.
Viseme template attribution
Section titled “Viseme template attribution”v0.3.0 uses MIT-licensed forward recipes from OpenFaceFX contributors (2026), revision f898db3c825bf89cfec63391bb16d91fc42192e8, and Mika Suominen / TalkingHead (2023–2024), revision 5b1f12057a0edad83d1fc75714217dbbc9496aa7. Recipe data, notices, and full licenses are in the plugin’s ThirdParty/VisemeTemplates. Keep these alongside the plugin when redistributing. The inverse fitter is original integration code; the recipes are not an inverse phoneme recognizer. Published notices