Add signed model bundle manifest for Desktop and ComfyUI

README.md

@@ -21,23 +21,33 @@ tags:
 
 # IndexTTS 2.5 · T8star-Aix 完整运行模型
 
-这是为 [T8mars/comfyui-indextts25-t8](https://github.com/T8mars/comfyui-indextts25-t8)
-整理的 **IndexTTS 2.5 完整运行目录**。下载本仓库后可直接放入：
+这是供以下两个项目共用的 **IndexTTS 2.5 完整推理模型包**：
 
-```text
-ComfyUI/models/TTS/IndexTTS-2.5/
-```
+- [T8star-Aix IndexTTS 2.5 Desktop](https://github.com/T8mars/indextts25-desktop-t8)
+- [T8star-Aix ComfyUI 节点](https://github.com/T8mars/comfyui-indextts25-t8)
+
+桌面版和 ComfyUI 使用的是同一套模型文件，无需分别下载。桌面启动器可以直接选择 ComfyUI 的
+`models/TTS/IndexTTS-2.5` 目录，共享一份约 7.72 GiB 的模型。
 
 本仓库解决官方 `IndexTeam/IndexTTS-2.5` 模型仓库未包含 `bpe.model`，以及首次运行还会从多个
-仓库下载 Wav2Vec2-BERT、CAMPPlus 和 BigVGAN 的问题。这里把节点实际推理需要的文件集中在一个
-目录中，方便 ComfyUI 用户一次下载和离线校验。
+仓库下载 Wav2Vec2-BERT、CAMPPlus 和 BigVGAN 的问题。这里把实际推理需要的 26 个文件集中在一个
+目录中，方便自动下载、断点续传、离线运行和 SHA-256 校验。
 
 > 本仓库不是 IndexTTS 官方发布渠道。所有权重保持源文件内容不变，仅重新整理目录。
 > 原始权利人不对此第三方整理仓库及其衍生集成提供认可、担保或保证，并对其不承担责任。
 
-## ComfyUI 节点与模型放置路径（重要）
+## 自动下载与模型版本
+
+仓库根目录的 `model-bundle.json` 是机器可读的完整文件清单，`model-bundle.sig` 是对应的
+Ed25519 签名。T8star-Aix 桌面启动器会先验证签名，再信任清单中的仓库版本、文件路径、大小和
+SHA-256；只下载缺失或损坏的文件，网络中断后可以继续下载。
 
-节点代码和模型是两个不同目录，请分别放置：
+模型更新由 `bundleVersion` 判断，不使用仓库 HEAD 判断。因此仅修改 README 不会误报模型更新。
+程序更新仍通过 GitHub Release 独立发布，模型不会被打进程序增量包。
+
+## ComfyUI 模型放置路径
+
+节点代码和模型是两个不同目录：
 
 ```text
 ComfyUI/
@@ -45,7 +55,7 @@ ComfyUI/
 │  └─ comfyui-indextts25-T8/       ← GitHub 节点代码
 └─ models/
    └─ TTS/
-      └─ IndexTTS-2.5/             ← 本 HF 仓库的全部模型文件
+      └─ IndexTTS-2.5/             ← 本 HF 仓库的模型文件
          ├─ config.yaml
          ├─ bpe.model
          ├─ gpt.pth
@@ -70,34 +80,41 @@ D:\ComfyUI\models\TTS\IndexTTS-2.5\bpe.model
 D:\ComfyUI\models\TTS\IndexTTS-2.5\hf_cache\bigvgan\bigvgan_generator.pt
 ```
 
-请确认 `config.yaml` 和 `bpe.model` 是 `IndexTTS-2.5` 的直接子文件。下面这种多套一层的结构是错误的：
+请确认 `config.yaml` 和 `bpe.model` 是 `IndexTTS-2.5` 的直接子文件。不要多套一层：
 
 ```text
 ComfyUI/models/TTS/IndexTTS-2.5/IndexTTS-2.5-Comfy/config.yaml  ← 错误
 ```
 
-如果浏览器下载解压后目录名是 `IndexTTS-2.5-Comfy`，请把它重命名为 `IndexTTS-2.5`，或把其中全部文件移动到标准目录。完成后重启或刷新 ComfyUI。
+## 桌面版模型放置路径
+
+桌面启动器支持两种方式：
 
-## 一键下载
+1. 点击“**Hugging Face 自动下载／修复完整模型**”，选择父目录，启动器会创建 `IndexTTS-2.5`；
+2. 选择已有完整目录，包括 ComfyUI 的 `ComfyUI/models/TTS/IndexTTS-2.5`。
+
+模型、保存音色、生成音频和用户设置都独立于桌面程序更新，更新应用不会删除这些内容。
+
+## 命令行下载
 
 ```powershell
 pip install -U huggingface_hub
 hf download t8star/IndexTTS-2.5-Comfy --local-dir "ComfyUI/models/TTS/IndexTTS-2.5"
 ```
 
-也可以在仓库页面选择 **Files and versions → Download**，解压后确保 `config.yaml`、`bpe.model`、
-`gpt.pth` 和 `hf_cache` 位于同一个 `IndexTTS-2.5` 目录下。
+也可以在仓库页面选择 **Files and versions → Download**。解压后确保 `config.yaml`、`bpe.model`、
+`gpt.pth` 和 `hf_cache` 位于同一个 `IndexTTS-2.5` 目录。
 
 ## 已包含内容
 
 - IndexTTS 2.5 主模型、语速适配模型、Qwen 文本情感模型与多语言分词文件
 - 官方 2.5 仓库缺失、但推理必需的 `bpe.model`
-- `hf_cache/w2v-bert-2.0`：运行所需的 Transformers safetensors、配置和预处理配置
+- `hf_cache/w2v-bert-2.0`：Transformers 权重、配置和预处理配置
 - `hf_cache/campplus_cn_common.bin`：音色特征模型
 - `hf_cache/bigvgan`：22.05 kHz BigVGAN 配置和生成器权重
 
 没有收录训练优化器、判别器、重复的 Fairseq `conformer_shaw.pt` 或当前 IndexTTS 2.5 推理代码
-不会读取的 MaskGCT 独立权重。这样可以保留完整推理能力，同时避免数 GB 的无用下载。
+不会读取的 MaskGCT 独立权重。
 
 ## 来源与固定版本
 
@@ -114,23 +131,13 @@ hf download t8star/IndexTTS-2.5-Comfy --local-dir "ComfyUI/models/TTS/IndexTTS-2
 
 ## English
 
-This is a single-directory, inference-only mirror prepared for the
-[T8star-Aix ComfyUI node](https://github.com/T8mars/comfyui-indextts25-t8). It combines the
-byte-identical official IndexTTS 2.5 files, the required `bpe.model` from the pinned IndexTTS 2
-repository, and the exact Wav2Vec2-BERT, CAMPPlus, and BigVGAN runtime files used by the node.
-
-Place the downloaded snapshot at `ComfyUI/models/TTS/IndexTTS-2.5/`. This is an unofficial
-packaging mirror and is not endorsed, warranted, or guaranteed by the original right-holders.
-Review `LICENSE` and `THIRD_PARTY_NOTICES.md` before downloading or using the files.
+This is the shared, inference-only IndexTTS 2.5 model bundle for the T8star-Aix Desktop package and
+ComfyUI node. Both applications use the same byte-identical files; the Desktop launcher can directly
+select `ComfyUI/models/TTS/IndexTTS-2.5`, so a second copy is unnecessary.
 
-Keep the node and model in separate directories:
-
-```text
-ComfyUI/custom_nodes/comfyui-indextts25-T8/   # GitHub node code
-ComfyUI/models/TTS/IndexTTS-2.5/             # all files from this HF repository
-```
+The repository combines the pinned official IndexTTS 2.5 files, the required `bpe.model`, and the exact
+Wav2Vec2-BERT, CAMPPlus, and BigVGAN runtime files. `model-bundle.json` and `model-bundle.sig` provide a
+signed, machine-readable version and per-file SHA-256 list for resumable automatic download and repair.
 
-`config.yaml`, `bpe.model`, and `gpt.pth` must be direct children of `IndexTTS-2.5`; `hf_cache`
-must remain inside the same directory. Avoid an extra nested folder such as
-`IndexTTS-2.5/IndexTTS-2.5-Comfy/config.yaml`. Rename the extracted folder or move its contents,
-then restart or refresh ComfyUI.
+This is an unofficial packaging mirror and is not endorsed, warranted, or guaranteed by the original
+right-holders. Review `LICENSE` and `THIRD_PARTY_NOTICES.md` before downloading or using the files.

model-bundle.json

@@ -0,0 +1,139 @@
+{
+  "schemaVersion": 1,
+  "bundleVersion": "1.0.0",
+  "publishedAt": "2026-08-29T00:00:00Z",
+  "minimumDesktopVersion": "0.19.1",
+  "minimumNodeVersion": "0.18.0",
+  "totalSize": 8290973742,
+  "codeRepository": "index-tts/index-tts",
+  "codeRevision": "ee40fa7d6c6b8a2c7f06105f9f1e65775b74868c",
+  "modelRepository": "t8star/IndexTTS-2.5-Comfy",
+  "modelRevision": "14166a7401f9f87f53770a1784390e8c0e9da15a",
+  "modelScopeRepository": "IndexTeam/IndexTTS-2.5",
+  "modelScopeRevision": "master",
+  "files": {
+    "config.yaml": {
+      "size": 2860,
+      "sha256": "18adf417be3e8f5e2e48e30f7420c719170a6870619436250f360d626877870e"
+    },
+    "bpe.model": {
+      "size": 475997,
+      "sha256": "b2a5ce8090d32da3642cc4f81fdc996376bc6dd3f4cd5e3d165f71120d9f2bc8",
+      "modelScopeRepository": "IndexTeam/IndexTTS-2"
+    },
+    "gpt.pth": {
+      "size": 3259599833,
+      "sha256": "43a8f4c30eccdf201958d3b9713511482c19d56dc20b0b1c4ee1e6b080b19d85"
+    },
+    "codec.pth": {
+      "size": 607290935,
+      "sha256": "d15cbed16a40f478438c961fb043f68dfa6353bf56c966761315db3433e9722c"
+    },
+    "s2mel.pth": {
+      "size": 414908601,
+      "sha256": "9b1b0003fc189c94cc349758d7ebc25f903b7eb2de4602879959cc64ce816456"
+    },
+    "feat1.pt": {
+      "size": 57170,
+      "sha256": "f219cb447d80216ba615666da2ff8d63ac544eee26657f3a7b278692bf7a67c4"
+    },
+    "feat2.pt": {
+      "size": 374866,
+      "sha256": "9c4292e96dee535aea9a6206e9a0c856dd578dde9212acdb16dd3ada4d12bf80"
+    },
+    "wav2vec2bert_stats.pt": {
+      "size": 9343,
+      "sha256": "c9c176c2b8850ab2e3ba828bbfa969deaf4566ce55db5f2687b8430b87526ad2"
+    },
+    "multilingual_zh_ja_yue_char_del.tiktoken": {
+      "size": 907395,
+      "sha256": "747979631e813193436aabcff7c1c235d37de8097b71c563ec8b63b7a515c718"
+    },
+    "qwen0.6bemo4-merge/model.safetensors": {
+      "size": 1192135096,
+      "sha256": "11293257a8df593c154a8ecd5fc039f3076de35411e35f06d41b471e136f6641"
+    },
+    "qwen0.6bemo4-merge/added_tokens.json": {
+      "size": 707,
+      "sha256": "c0284b582e14987fbd3d5a2cb2bd139084371ed9acbae488829a1c900833c680"
+    },
+    "qwen0.6bemo4-merge/chat_template.jinja": {
+      "size": 550,
+      "sha256": "a13337ea164ac47b2c1e644b6338d5e7e91cdcb58218511899c083112472730c"
+    },
+    "qwen0.6bemo4-merge/generation_config.json": {
+      "size": 117,
+      "sha256": "6453260cdb41552215d87400ab399a760f25509ec2860ceb07817b195651626d"
+    },
+    "qwen0.6bemo4-merge/merges.txt": {
+      "size": 1671853,
+      "sha256": "8831e4f1a044471340f7c0a83d7bd71306a5b867e95fd870f74d0c5308a904d5"
+    },
+    "qwen0.6bemo4-merge/Modelfile": {
+      "size": 360,
+      "sha256": "986c18655a26bbbdeb486a9478c8653925c7db7111d6cf20c9991b348853cb95"
+    },
+    "qwen0.6bemo4-merge/special_tokens_map.json": {
+      "size": 616,
+      "sha256": "6676f091c8bc4d1b50146427cfde92073402866b87b6e39223227931b70083e9"
+    },
+    "qwen0.6bemo4-merge/tokenizer.json": {
+      "size": 11422654,
+      "sha256": "aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4"
+    },
+    "qwen0.6bemo4-merge/tokenizer_config.json": {
+      "size": 5433,
+      "sha256": "147bae7844ddfb8767c1f9cc0f505d4a26839bd6fbb0656d6d6bc17c82e9108a"
+    },
+    "qwen0.6bemo4-merge/vocab.json": {
+      "size": 2776833,
+      "sha256": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910"
+    },
+    "qwen0.6bemo4-merge/config.json": {
+      "size": 727,
+      "sha256": "2e04bc223dfb548dde52faebe2a819964523ff749cbe26467c49380aa371594f"
+    },
+    "hf_cache/w2v-bert-2.0/config.json": {
+      "size": 1874,
+      "sha256": "f5572bd5998b68182e9c328a43127ed21fed687f6910497136b91a4e3b0e3675",
+      "group": "auxiliary",
+      "sourceRepository": "facebook/w2v-bert-2.0",
+      "sourceRevision": "da985ba0987f70aaeb84a80f2851cfac8c697a7b"
+    },
+    "hf_cache/w2v-bert-2.0/model.safetensors": {
+      "size": 2322063736,
+      "sha256": "eb890c9660ed6e3414b6812e27257b8ce5454365d5490d3ad581ea60b93be043",
+      "group": "auxiliary",
+      "sourceRepository": "facebook/w2v-bert-2.0",
+      "sourceRevision": "da985ba0987f70aaeb84a80f2851cfac8c697a7b"
+    },
+    "hf_cache/w2v-bert-2.0/preprocessor_config.json": {
+      "size": 275,
+      "sha256": "8e6281aad64f97e40534135a59dcc5d33571efae376f2a25adf5551951897ab4",
+      "group": "auxiliary",
+      "sourceRepository": "facebook/w2v-bert-2.0",
+      "sourceRevision": "da985ba0987f70aaeb84a80f2851cfac8c697a7b"
+    },
+    "hf_cache/campplus_cn_common.bin": {
+      "size": 28036335,
+      "sha256": "3388cf5fd3493c9ac9c69851d8e7a8badcfb4f3dc631020c4961371646d5ada8",
+      "group": "auxiliary",
+      "sourceRepository": "funasr/campplus",
+      "sourceRevision": "e4b6ede7ce16997aff4ae69fbca1f0175e2afede"
+    },
+    "hf_cache/bigvgan/config.json": {
+      "size": 1405,
+      "sha256": "88a1f47acf747db0b21e97a389d838566147f7a5464583ff5c8d819d870f03ee",
+      "group": "auxiliary",
+      "sourceRepository": "nvidia/bigvgan_v2_22khz_80band_256x",
+      "sourceRevision": "633ff708ed5b74903e86ff1298cf4a98e921c513"
+    },
+    "hf_cache/bigvgan/bigvgan_generator.pt": {
+      "size": 449228171,
+      "sha256": "e95ba25972d3de0628d99cd156e9315a9c018899bf739988959ebe3544080ced",
+      "group": "auxiliary",
+      "sourceRepository": "nvidia/bigvgan_v2_22khz_80band_256x",
+      "sourceRevision": "633ff708ed5b74903e86ff1298cf4a98e921c513"
+    }
+  }
+}

model-bundle.sig

@@ -0,0 +1 @@
+Epe8cSV8c8J9QG/WlmLK3LUBvP0LhqUM2JziSbr88SFFFSGLdXP9UEa+SaMwsOKO1PCitVf/LboSDKCfvcocAA==