Style-Bert-VITS2 (litagin02): Bert-VITS2 with controllable voice styles, Japanese TTS and training UI
Source: https://github.com/litagin02/style-bert-vits2
Pretrained base model record: hf--litagin--style-bert-vits2-2-0-base-jp-extra. Default voice models bundled at install time carry their own terms (JVNV corpus CC BY-SA 4.0; あみたろ/小春音アミ with credit requirement) per docs/TERMS_OF_USE.md.
Structured fields
{
"archived": false,
"revision": {
"ref": "2.7.0 (latest release, published 2025-08-24)",
"pinned": true,
"source_url": "https://github.com/litagin02/Style-Bert-VITS2/releases"
},
"checked_at": "2026-10-17T00:00:00.000Z",
"parent_url": "https://github.com/fishaudio/Bert-VITS2",
"usefulness": "Japanese-first expressive TTS built on Bert-VITS2 v2.1 and JP-Extra, with style vectors, model merging, ONNX export, a training WebUI and an API server. Runs synthesis and merging on CPU and trains on NVIDIA GPUs; code is AGPL-3.0, and the README plus a terms file add model-level obligations (CC BY-SA 4.0 for the default JVNV-based voices, credit lines for the Amitaro model).",
"source_type": "repository",
"availability": "available",
"dependencies": [
"Python",
"PyTorch",
"Bert-VITS2 lineage",
"pip package style-bert-vits2 for inference only"
],
"source_owner": "litagin02 (individual developer; Japanese-language README)",
"next_check_at": "2026-10-31T00:00:00.000Z",
"hardware_notes": "README: synthesis works on CPU without a GPU; training needs an NVIDIA GPU; merging and synthesis work without one; CPU install script provided. No VRAM numbers read.",
"known_limitations": [
"AGPL-3.0 network-copyleft applies to code; the repository also carries LGPL-3.0 text for text/user_dict derived from VOICEVOX engine.",
"The terms file separates non-binding requests (\"お願い\") from binding model terms; only the model terms and the repository license bind.",
"The pretrained base weights are derived from Bert-VITS2 (Fish Audio) models and are labelled AGPL-3.0 on Hugging Face; provenance and licenses of the training data of those base models were not traced.",
"The owner's location was not verified; the project is Japan-focused and documented in Japanese."
],
"upstream_updated_at": "2025-12-07T13:06:59.000Z"
}Evidence claims
- [sourced_fact] The README says the project modifies Bert-VITS2 v2.1 and Japanese-Extra to let users control emotion and speaking style, and that synthesis alone works on CPU without a graphics card. · locator: README, 概要 source
- [sourced_fact] Without an NVIDIA GPU, training is not possible but synthesis and model merging still work. · locator: README, 動作環境 source
- [sourced_fact] The repository is licensed under AGPL-3.0, the same as Bert-VITS2, with text/user_dict under LGPL v3. · locator: README, LICENSE source
- [sourced_fact] The default JVNV-corpus-based models (jvnv-F1-jp, F2, M1, M2) inherit the corpus license CC BY-SA 4.0. · locator: docs/TERMS_OF_USE.md, モデルの利用規約・ライセンス, JVNVコーパス source
- [sourced_fact] For the 小春音アミ/あみたろ models, published generated audio must carry a credit that the voice is based on あみたろの声素材工房, with use cases such as age-restricted works and treating output as the speaker's own voice prohibited. · locator: docs/TERMS_OF_USE.md, 小春音アミ / あみたろ source
- [sourced_fact] The terms file says the sections before the model terms are non-binding requests (not prohibitions) and that only the repository license binds for repository use; requests include no deepfake/impersonation use. · locator: docs/TERMS_OF_USE.md, header source
- [sourced_fact] GitHub API returns the repository record for litagin02/Style-Bert-VITS2: default branch master; pushed_at at check time 2025-12-07T13:06:59Z (raw response saved under evidence/github/litagin02--style-bert-vits2.json). · locator: GitHub API repos/litagin02/Style-Bert-VITS2, default_branch (pushed_at is volatile) source
- [sourced_fact] GitHub API reports archived = false for litagin02/Style-Bert-VITS2. · locator: GitHub API repos/litagin02/Style-Bert-VITS2, archived source
- [sourced_fact] GitHub license detection for litagin02/Style-Bert-VITS2 returns spdx_id AGPL-3.0. · locator: GitHub API repos/litagin02/Style-Bert-VITS2, license source
- [sourced_fact] GitHub API latest release for litagin02/Style-Bert-VITS2: tag 2.7.0 published 2025-08-24T03:01:47Z (selected fields of the response saved under evidence/github/litagin02--style-bert-vits2.release.json). · locator: GitHub API repos/litagin02/Style-Bert-VITS2/releases/latest, tag_name source
- [upstream_claim] Research review verdict at export: reviewed (as of 2026-10-17). Recorded by the research pipeline, not a hub review outcome; evidence.review stays unreviewed until an independent reviewer records one.
Rights
| component | license | hosting | redistribution | modification | training | Evidence and notes |
|---|---|---|---|---|---|---|
| code | AGPL-3.0-only | permitted | restricted | restricted | unknown | AGPL-3.0 with LGPL-3.0 for text/user_dict. Network-use source-offer duty; GitHub detects AGPL-3.0. Does not address training. |
| model_weights | LicenseRef-per-model-terms | restricted | restricted | restricted | unknown | Not the repository license: base weights are labelled AGPL-3.0 on Hugging Face; default JVNV-based voices inherit CC BY-SA 4.0 (ShareAlike); the Amitaro voice models require a credit and forbid listed uses and have merge limits (25 percent or more Amitaro share keeps their terms). Check each model's terms. |
| other | LicenseRef-per-model-terms | not_applicable | not_applicable | not_applicable | unknown | outputs: credit lines are required when publishing audio from the Amitaro-based models (example "Style-BertVITS2モデル: 小春音アミ、あみたろの声素材工房 (https://amitaro.net/)"); the JVNV models carry CC BY-SA 4.0 obligations; project requests ask not to use for impersonation or deepfakes. |