Confucius4-R2T2 Q8 for ComfyUI
English
This repository mirrors the unmodified official Q8 GGUF decoder and Q8 audio projector from NetEase Youdao, plus the FireRedVAD ONNX assets used by Confucius4-R2T2 Q8 for ComfyUI. It is not an official NetEase Youdao or FireRedVAD release.
| File | SHA-256 | Source |
|---|---|---|
Confucius4-R2T2-Q8_0.gguf |
151097e43957a19984ea7de66e8144ce69b95039eb31c93da4f58db367e455c3 |
NetEase Youdao GGUF revision 86ff0251cb9f456b63aeef5f80137f104e22869a |
mmproj-Confucius4-R2T2-Q8_0.gguf |
8dc2c67e6a0484114928142d098db7ad94ae9f34c78948ef9d37a9678418cb65 |
Same revision |
FireRedVAD-ONNX/fireredvad_stream_vad_with_cache.onnx |
b3c97836130dc34fc32d56fab551e88cf9454511de2b3c250a5e6578dee74b93 |
FireRedVAD commit c30ec49e8cc69642b0ee65362eba11b9d11c6e54 |
FireRedVAD-ONNX/cmvn.ark |
c87f6f13edf0f0ec7535ddfc9cc3387d9268cb234b70182d566c5e2edf3ca473 |
Same commit |
The NetEase Youdao files are governed by the included Model Use License Agreement (Chinese), not by the FireRedVAD license. The VAD files are governed by Apache 2.0. Keep the applicable notices and license terms when redistributing. The ComfyUI setup script verifies exact file sizes and SHA-256 hashes before inference.
简体中文
本仓库镜像网易有道未经修改的官方 Q8 GGUF 主模型与 Q8 音频 projector，并提供 ComfyUI 节点 使用的 FireRedVAD ONNX 文件。本仓库不是网易有道或 FireRedVAD 官方发布。文件来源、固定版本与 SHA-256 见上表。GGUF 文件适用随附的网易有道模型许可；VAD 文件适用 Apache 2.0。使用或再分发前请阅读并保留相应许可。
T8star-Aix Voice Studio Windows integration / Windows 桌面整合包
The Breeze desktop bundle 0.3.8 reuses the existing pinned Confucius worker alongside Whisper Large-v3. Both the generation page and the voice library offer an ASR engine selector. No ComfyUI node update or vLLM installation is required. The full portable bundle includes the unchanged Q8 pair, FireRedVAD, a separate portable Python 3.12 worker and its CUDA runtime; an NVIDIA driver/GPU is still required. This documentation update does not re-upload or change any model weights.
Windows 完整整合包默认目录：<解压目录>\resources\backend\models\Confucius4-R2T2-GGUF。在 设置与诊断 → 转录引擎与模型路径 选择并保存自定义目录（跨重启有效），或恢复内置目录。所选目录直接包含上表两个 GGUF 文件；可在其中附 FireRedVAD-ONNX/，否则复用整合包内置 VAD。可以直接复用已有节点的模型目录，不必重复占用磁盘。
Confucius 热词与识别上下文可在设置页填写，生成页/音色库共用。转录只生成需要人工逐字核对的完整草稿，不改参考录音，不是翻译，也不伪造字幕时间戳；需要时间范围可选 Whisper。转录与 TTS 串行使用 GPU，结束后关闭识别 worker 释放显存。
Desktop releases / 整合包下载 · Breeze 模型镜像与路径说明 · Source pins and licenses / 源码与许可
Links / 链接
GitHub / ComfyUI node · Bilibili / B站 · YouTube · API · Free gallery / 免费画廊 · Online AI apps / 在线 AI 应用 · ComfyUI package / 整合包 · Hugging Face
- Downloads last month
- 1,240
8-bit