It’s a bit of a shame that my V100 won’t work with Jasna.

[pollux94594]

[qiyu55555]

请确保cuda版本在13.0以上，这是硬性要求

[pollux94594]

请确保cuda版本在13.0以上，这是硬性要求

我肯定不会犯这么低级的错误 应该是V100无法兼容了

[Kruk2]

Due to my limited time I can only provide prebuild packages for cuda 13.0
If someone else wants to provide other versions feel free to do so. Or try to setup github pipeline to build them automatically.

AFAIK there is nothing that should stop you from using cuda 12.9. You need to install pytorch in correct version and build two packages listed in readme with your env.
I don't know what encoder capabilities V100 has so you need to investigate it because it might be the problem. Currently Jasna requires hevc with bframes support (but b-frames can be disabled via custom encoder settings).

[pollux94594]

Could you provide the method for compiling and installing from the Windows source code? I want to adjust the PyTorch version on my own for experimentation.

[Kruk2]

I think there is a guide in readme, on the bottom.
Not easy since you need to compile two c++ programs and have cmake but doable (especially if someone knows how to use AI)

[pollux94594]

I have successfully compiled and installed PyNvVideoCodec, but I have no idea how to use it. It seems that the models and other related files are not included in it, and there is no executable EXE program either.

[pollux94594]

我已经成功编译并安装了 PyNvVideoCodec，但我不知道该如何使用它。似乎模型和其他相关文件都没有包含在内，也没有可执行的 EXE 程序。

1. 安装PyTorch（适配CUDA13.0，原命令cu128/cu126均兼容，保留其一即可）
   pip3 install torch==2.10.0 torchvision==0.25.0 torchaudio==2.10.0 --index-url https://download.pytorch.org/whl/cu128 -i https://pypi.tuna.tsinghua.edu.cn/simple

2. 全局安装uv包管理器（替代pip，提升编译/安装速度）
   pip3 install uv -i https://pypi.tuna.tsinghua.edu.cn/simple

3. 全局安装编译核心依赖（cmake+ninja，Kruk2仓库编译必备）
   uv pip install cmake ninja -i https://pypi.tuna.tsinghua.edu.cn/simple --system

4. 克隆Jasna主项目（核心仓库）
   git clone https://github.com/Kruk2/jasna.git
   cd jasna

5. 创建3rdparty目录，存放Vali/PyNvVideoCodec依赖
   mkdir 3rdparty && cd 3rdparty

6. 克隆Kruk2定制版Vali（核心视频编解码依赖）
   git clone https://codeberg.org/Kruk2/vali.git

7. 克隆Kruk2定制版PyNvVideoCodec（基于Vali的Python封装）
   git clone https://codeberg.org/Kruk2/PyNvVideoCodec.git

8. 进入Vali目录，初始化并更新子模块（修复dlpack/pybind11等依赖缺失）
   cd vali
   git submodule sync --recursive
   git submodule update --init --recursive --force

9. 创建build目录，分离源码与编译文件（CMake标准流程）
   mkdir build && cd build

10. CMake配置（简化命令，Kruk2/Vali无需手动指定FFmpeg参数，自动识别）
    cmake .. -G Ninja ^
    -DCMAKE_C_COMPILER=cl.exe ^
    -DCMAKE_CXX_COMPILER=cl.exe ^
    -DCMAKE_BUILD_TYPE=Release ^
    -DPYTHON_EXECUTABLE=C:\Users\Administrator\AppData\Local\Programs\Python\Python313\python.exe

11. 编译Vali（Ninja高效构建，基于VS2022+CUDA13.0）
    cmake --build . --config Release

12. 安装Vali到Python3.13全局环境（核心步骤，让Python识别Vali）
    cmake --install . --config Release --prefix C:\Users\Administrator\AppData\Local\Programs\Python\Python313

13. 返回3rdparty目录，进入Kruk2/PyNvVideoCodec目录
    cd ../../PyNvVideoCodec

14. 编译安装PyNvVideoCodec（复用系统环境，无构建隔离，依赖已安装的Vali/CUDA）
    uv pip install . --no-build-isolation --system -i https://pypi.tuna.tsinghua.edu.cn/simple

15. 返回Jasna主目录（核心，准备安装Jasna开发版）
    cd ../../

16. 安装scikit-build（Jasna编译必备的Python构建工具，补全依赖）
    uv pip install scikit-build --system -i https://pypi.tuna.tsinghua.edu.cn/simple

17. 安装Jasna可编辑开发版（-e .[dev]，修改源码无需重新安装，包含开发依赖）
    uv pip install -e .[dev] --system -i https://pypi.tuna.tsinghua.edu.cn/simple

[ASMR-Seduction]

It is indeed quite complex to run, and not suitable for beginners. It is recommended to install a one-click startup program.