已支持30系以下显卡和首尾帧 已支持api调用 已支持lora和文生视频
Support for graphics cards below the 30-series and for first/last frames has been added. API calls are now supported. Lora and text-to-video generation are now supported.
python -m scripts.app
或者双击 单独启动.bat
Or double-click 单独启动.bat (Standalone Startup.bat)
注意,直接运行不要运行 install.py ,而是先装 torch和 torchvision ,然后 pip install -r requirements.txt 把必要依赖装好,然后建议 xformers (配套装triton,windows安装方法上网查),flash attention 或者 sage attention 装上,否则会很慢
Note: For direct execution, do not run install.py. Instead, first install torch and torchvision. Then, run pip install -r requirements.txt to install the necessary dependencies. It is also recommended to install xformers (along with triton; search online for Windows installation methods), flash attention, or sage attention, otherwise it will be very slow.
2. Place it in the webui's extensions folder. The model will be automatically downloaded to models/hunyuan. If you have lora models, place them in the models/hunyuan/lora folder.
根据测试,冲突的环境有这些: According to tests, these are the conflicting environments:
Installed version of accelerate (0.31.0) does not satisfy the requirement (==1.6.0).
Installing accelerate==1.6.0
Installed version of diffusers (0.31.0) does not satisfy the requirement (>=0.33.1).
Installing diffusers from source
Installed version of pillow (10.4.0) does not satisfy the requirement (==11.1.0).
Installing pillow==11.1.0
解决方法是去你的webui根目录下,找到requirements_versions.txt,把里面相关的冲突依赖删掉,有效避免每次启动都安装依赖
The solution is to go to your webui root directory, find requirements_versions.txt, and delete the conflicting dependencies listed within. This will effectively prevent the installation of dependencies every time you start.
不过如果要省事你可以在启动成功后加上启动参数 --skip-prepare-environment ,这样下次启动就不用依赖安装浪费一堆时间了
However, if you want to save trouble, you can add the launch parameter --skip-prepare-environment after a successful startup. This way, you won't waste a lot of time on dependency installation the next time you start.