尧图网页设计 
前置说明与目标显卡Tesla V100 16GSXM 自改成PCIe 价格实惠cup型号 E5 2673V3主板X99 P8D3内存大小 2*4g 2*8g 24g ddr3系统Ubuntu 24.04目标跑通 Wan2.2 14B GGUF 量化版文生视频输出一段 3 秒 640×640 视频还会泡泡14B fp8 scaled.预计首次全流程耗时模型下载 180分钟取决于网速第一步系统环境准备1、按照ubuntu24.04这个教材很多不啰嗦自己找来装装server版本个人习惯2、更新系统和软件包sudo apt updatesudo apt upgradesudo apt dist-upgradesudo apt autoremove3、配置swap大小40G大概内存的1.5到2倍4、禁用nouveau编辑文件blacklist.confsudo nano /etc/modprobe.d/blacklist.conf在文件最后部分插入以下两行内容blacklist nouveauoptions nouveau modeset0更新系统sudo update-initramfs -u重启系统一定要重启sudo reboot验证nouveau是否已禁用lsmod | grep nouveau没有信息显示说明nouveau已被禁用接下来可以安装nvidia的显卡驱动。5、安装基础工具sudo apt install -y wget git curl python3-pip python3-dev gcc g make6、安装驱动检查显卡需要亮集卡我这里是已经装好了所以亮机卡拿掉了(base) jjjjjj:~$ sudo lshw -class display[sudo] jjj 的密码*-displaydescription: 3D controllerproduct: GV100GL [Tesla V100 SXM2 16GB]vendor: NVIDIA Corporationphysical id: 0bus info: pci0000:03:00.0version: a1width: 64 bitsclock: 33MHzcapabilities: pm msi pciexpress bus_master cap_listconfiguration: drivernvidia latency0resources: iomemory:38380-3837f iomemory:383c0-383bf irq:62 memory:fa000000-faffffff memory:383800000000-383bffffffff memory:383c00000000-383c01ffffff查找可安装的版本(base) jjjjjj:~$ ubuntu-drivers devicesvendor : NVIDIA Corporationmodel : GV100GL [Tesla V100 SXM2 16GB]driver : nvidia-driver-470-server - distro non-freedriver : nvidia-driver-580 - distro non-free recommendeddriver : nvidia-driver-580-server - distro non-freedriver : xserver-xorg-video-nouveau - distro free builtin安装推荐的驱动sudo apt installnvidia-driver-580-y#安装完成需要重启sudo reboot检查一下安装结果(base) jjjjjj:~$ nvidia-smiMon Oct 5 20:08:30 2026-----------------------------------------------------------------------------------------|NVIDIA-SMI 580.178.04 Driver Version: 580.178.04 CUDA Version: 13.0|---------------------------------------------------------------------------------------| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC || Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. || | | MIG M. |||| 0 Tesla V100-SXM2-16GB On | 00000000:03:00.0 Off | 0 || N/A 29C P0 24W / 300W | 0MiB /16384MiB| 0% Default || | | N/A |--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| Processes: || GPU GI CI PID Type Process name GPU Memory || ID ID Usage |||| No running processes found |-----------------------------------------------------------------------------------------这样我们驱动就安装完成了7、按照cuda可以不装我记得不编译不装也是可以的#12.6wget https://developer.download.nvidia.com/compute/cuda/12.6.3/local_installers/cuda_12.6.3_560.35.05_linux.runsudo sh cuda_12.6.3_560.35.05_linux.runecho export PATH/usr/local/cuda-12.6/bin:PATH′ /.bashrcecho′exportLDLIBRARYPATH/usr/local/cuda−12.6/lib64:LD_LIBRARY_PATH ~/.bashrcsource ~/.bashrcnvidia-smi # 应能看到 V100 和 CUDA 版本nvcc --version # 应显示 CUDA 版本8、安装Miniconda3,创建虚拟环境安装Miniconda3#下载wget -c https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh#安装bash Miniconda3-latest-Linux-x86_64.sh#初始化source ~/.bashrc#检验(base) jjjjjj:~$ conda --versionconda 26.7.1创建虚拟环境#创建环境conda create -n comfyui python3.12 -y#激活环境conda activate comfyui#升级python工具也可不进行python -m pip install -U pip wheel setuptools#安装ModelScopepip install modelscope配置清华源速度不慢不用设置设置了可能会有新的问题pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple安装pytorch,安装cuda12.6#12.6pip install torch2.13.0 torchvision0.28.0 torchaudio2.11.0 --index-url https://download.pytorch.org/whl/cu126测试python -c import torch; print(torch.__version__)python -c import torch; print(torch.version.cuda)python -c import torch; print(torch.cuda.get_device_name(0) if torch.cuda.is_available() else CUDA is not available)第二步配置Comfyui1、下载 ComfyUI并安装cd ~git clone https://github.com/Comfy-Org/ComfyUI.gitcd ~/ComfyUIpip install -r requirements.txt如果不能下载可以给你们个代理不一定看到这篇文章时还能用GitHub Proxy - GitHub 文件下载加速代理ERROR: Could not find a version that satisfies the requirement comfyui-workflow-templates-media-assets-020.1.5 (from comfyui-workflow-templates) (from versions: none)ERROR: No matching distribution found for comfyui-workflow-templates-media-assets-020.1.5如果出现上面错误pip config unset global.index-url就是设置了清华原没有最新的ComfyUI 的核心依赖包括 comfyui-workflow-templates经常更新这里就有一个问题因为comfyui更新很平凡所以有些依赖包镜像站点来不及更新所以不要设置就是太慢了2、安装 ComfyUI Managercd ~/ComfyUIpip install -r manager_requirements.txt3、启动 ComfyUI#有错误免得一个一个来先安装有怀疑的可以不装后面会报错但是版本再更新不好说建议还是第一次一个一个来吧写在这里是给我自己参考的pip install matrix-nio PyOpenGL PyOpenGL-accelerate pynvmlcd ~/ComfyUIconda activate comfyuipython /home/jjj/ComfyUI/main.py --listen 0.0.0.0 --port 8188 --fast fp16_accumulation --enable-manager --output-directory /mnt/data_160g/ComfyUI_Outputs我把输出放在了机械硬盘实在是模型太大了关于参数我也抄的没来及想4、浏览器访问http://192.168.1.22:81885、安装必要插件大家不用怀疑直接装吧安装有两种方法第一种下载源码放在custom-node 目录下面目录下有requirements.txt文件的pip install -r requirements.txt然后重新运行comfyui重启浏览器第二种在管理扩展功能中安装如下图这个功能和别的软件插件功能一样没什么好研究的直接安装就是如果这里不能安装就用第一种目前为止我装了9个插件6、安装ComfyUI-WanVideoWrapperWan 专用节点我们要用wan2.2生成视频所以要安装这个插件不过新的版本已经集成了新的wan,所以不用装我装了没成功啊有了就不纠结第三步加载官方模板首次出片1、下载模型平时在这里找国内好使但是不一定在这里下载我们直接使用Comfyui自带模板演示需要下载的模型目录ComfyUI/ ├─── models/ │ ├─── diffusion_models/ │ │ ├─── wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors │ │ └─── wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors │ ├─── loras/ │ │ ├───wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensors │ │ └───wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors │ ├─── text_encoders/ │ │ └─── umt5_xxl_fp8_e4m3fn_scaled.safetensors │ └─── vae/ │ └── wan_2.1_vae.safetensors本篇文章只是为了演示不重点阐述点击下载会显示下载地址https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors修改为https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors我是ubuntu系统直接命令行wget -P /home/jjj/ComfyUI/models/loras https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors wget -P /home/jjj/ComfyUI/models/loras https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensors wget -P /home/jjj/ComfyUI/models/text_encoders https://hf-mirror.com/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors wget -P /home/jjj/ComfyUI/models/diffusion_models https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors wget -P /home/jjj/ComfyUI/models/diffusion_models https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors wget -P /home/jjj/ComfyUI/models/vae https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors等候下载。。。。。。。。。。。大概2小时我家网速慢华数的9MB/s的速度说实话很快了就怪模型太大在家也不干什么看着便宜就他划算老小区没光纤没法升级下车要换移动和电信了2、运行我们都用了默认的参数只是演示一下跑完了,[INFO] Prompt executed in 00:29:12时间够长我们打开enable_turbo_mode,现在不用搞清楚什么意思就看能不能跑与lora加速有关[INFO] Prompt executed in 190.14 seconds效果明显当然这个有第二次运行可能缓存里已经有数据有关。大家可以试试清楚缓存或重启comfyui后续说明下一篇我们将详细解释节点参数并且实验多种模型已经说明他们的区别 【ComfyUI 国产模型炼丹记】第一篇：V100 16G 部署 Wan2.2：从环境配置到首次出片
前置说明与目标显卡Tesla V100 16GSXM 自改成PCIe 价格实惠cup型号 E5 2673V3主板X99 P8D3内存大小 2*4g 2*8g 24g ddr3系统Ubuntu 24.04目标跑通 Wan2.2 14B GGUF 量化版文生视频输出一段 3 秒 640×640 视频还会泡泡14B fp8 scaled.预计首次全流程耗时模型下载 180分钟取决于网速第一步系统环境准备1、按照ubuntu24.04这个教材很多不啰嗦自己找来装装server版本个人习惯2、更新系统和软件包sudo apt updatesudo apt upgradesudo apt dist-upgradesudo apt autoremove3、配置swap大小40G大概内存的1.5到2倍4、禁用nouveau编辑文件blacklist.confsudo nano /etc/modprobe.d/blacklist.conf在文件最后部分插入以下两行内容blacklist nouveauoptions nouveau modeset0更新系统sudo update-initramfs -u重启系统一定要重启sudo reboot验证nouveau是否已禁用lsmod | grep nouveau没有信息显示说明nouveau已被禁用接下来可以安装nvidia的显卡驱动。5、安装基础工具sudo apt install -y wget git curl python3-pip python3-dev gcc g make6、安装驱动检查显卡需要亮集卡我这里是已经装好了所以亮机卡拿掉了(base) jjjjjj:~$ sudo lshw -class display[sudo] jjj 的密码*-displaydescription: 3D controllerproduct: GV100GL [Tesla V100 SXM2 16GB]vendor: NVIDIA Corporationphysical id: 0bus info: pci0000:03:00.0version: a1width: 64 bitsclock: 33MHzcapabilities: pm msi pciexpress bus_master cap_listconfiguration: drivernvidia latency0resources: iomemory:38380-3837f iomemory:383c0-383bf irq:62 memory:fa000000-faffffff memory:383800000000-383bffffffff memory:383c00000000-383c01ffffff查找可安装的版本(base) jjjjjj:~$ ubuntu-drivers devicesvendor : NVIDIA Corporationmodel : GV100GL [Tesla V100 SXM2 16GB]driver : nvidia-driver-470-server - distro non-freedriver : nvidia-driver-580 - distro non-free recommendeddriver : nvidia-driver-580-server - distro non-freedriver : xserver-xorg-video-nouveau - distro free builtin安装推荐的驱动sudo apt installnvidia-driver-580-y#安装完成需要重启sudo reboot检查一下安装结果(base) jjjjjj:~$ nvidia-smiMon Oct 5 20:08:30 2026-----------------------------------------------------------------------------------------|NVIDIA-SMI 580.178.04 Driver Version: 580.178.04 CUDA Version: 13.0|---------------------------------------------------------------------------------------| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC || Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. || | | MIG M. |||| 0 Tesla V100-SXM2-16GB On | 00000000:03:00.0 Off | 0 || N/A 29C P0 24W / 300W | 0MiB /16384MiB| 0% Default || | | N/A |--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| Processes: || GPU GI CI PID Type Process name GPU Memory || ID ID Usage |||| No running processes found |-----------------------------------------------------------------------------------------这样我们驱动就安装完成了7、按照cuda可以不装我记得不编译不装也是可以的#12.6wget https://developer.download.nvidia.com/compute/cuda/12.6.3/local_installers/cuda_12.6.3_560.35.05_linux.runsudo sh cuda_12.6.3_560.35.05_linux.runecho export PATH/usr/local/cuda-12.6/bin:PATH′ /.bashrcecho′exportLDLIBRARYPATH/usr/local/cuda−12.6/lib64:LD_LIBRARY_PATH ~/.bashrcsource ~/.bashrcnvidia-smi # 应能看到 V100 和 CUDA 版本nvcc --version # 应显示 CUDA 版本8、安装Miniconda3,创建虚拟环境安装Miniconda3#下载wget -c https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh#安装bash Miniconda3-latest-Linux-x86_64.sh#初始化source ~/.bashrc#检验(base) jjjjjj:~$ conda --versionconda 26.7.1创建虚拟环境#创建环境conda create -n comfyui python3.12 -y#激活环境conda activate comfyui#升级python工具也可不进行python -m pip install -U pip wheel setuptools#安装ModelScopepip install modelscope配置清华源速度不慢不用设置设置了可能会有新的问题pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple安装pytorch,安装cuda12.6#12.6pip install torch2.13.0 torchvision0.28.0 torchaudio2.11.0 --index-url https://download.pytorch.org/whl/cu126测试python -c import torch; print(torch.__version__)python -c import torch; print(torch.version.cuda)python -c import torch; print(torch.cuda.get_device_name(0) if torch.cuda.is_available() else CUDA is not available)第二步配置Comfyui1、下载 ComfyUI并安装cd ~git clone https://github.com/Comfy-Org/ComfyUI.gitcd ~/ComfyUIpip install -r requirements.txt如果不能下载可以给你们个代理不一定看到这篇文章时还能用GitHub Proxy - GitHub 文件下载加速代理ERROR: Could not find a version that satisfies the requirement comfyui-workflow-templates-media-assets-020.1.5 (from comfyui-workflow-templates) (from versions: none)ERROR: No matching distribution found for comfyui-workflow-templates-media-assets-020.1.5如果出现上面错误pip config unset global.index-url就是设置了清华原没有最新的ComfyUI 的核心依赖包括 comfyui-workflow-templates经常更新这里就有一个问题因为comfyui更新很平凡所以有些依赖包镜像站点来不及更新所以不要设置就是太慢了2、安装 ComfyUI Managercd ~/ComfyUIpip install -r manager_requirements.txt3、启动 ComfyUI#有错误免得一个一个来先安装有怀疑的可以不装后面会报错但是版本再更新不好说建议还是第一次一个一个来吧写在这里是给我自己参考的pip install matrix-nio PyOpenGL PyOpenGL-accelerate pynvmlcd ~/ComfyUIconda activate comfyuipython /home/jjj/ComfyUI/main.py --listen 0.0.0.0 --port 8188 --fast fp16_accumulation --enable-manager --output-directory /mnt/data_160g/ComfyUI_Outputs我把输出放在了机械硬盘实在是模型太大了关于参数我也抄的没来及想4、浏览器访问http://192.168.1.22:81885、安装必要插件大家不用怀疑直接装吧安装有两种方法第一种下载源码放在custom-node 目录下面目录下有requirements.txt文件的pip install -r requirements.txt然后重新运行comfyui重启浏览器第二种在管理扩展功能中安装如下图这个功能和别的软件插件功能一样没什么好研究的直接安装就是如果这里不能安装就用第一种目前为止我装了9个插件6、安装ComfyUI-WanVideoWrapperWan 专用节点我们要用wan2.2生成视频所以要安装这个插件不过新的版本已经集成了新的wan,所以不用装我装了没成功啊有了就不纠结第三步加载官方模板首次出片1、下载模型平时在这里找国内好使但是不一定在这里下载我们直接使用Comfyui自带模板演示需要下载的模型目录ComfyUI/ ├─── models/ │ ├─── diffusion_models/ │ │ ├─── wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors │ │ └─── wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors │ ├─── loras/ │ │ ├───wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensors │ │ └───wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors │ ├─── text_encoders/ │ │ └─── umt5_xxl_fp8_e4m3fn_scaled.safetensors │ └─── vae/ │ └── wan_2.1_vae.safetensors本篇文章只是为了演示不重点阐述点击下载会显示下载地址https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors修改为https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors我是ubuntu系统直接命令行wget -P /home/jjj/ComfyUI/models/loras https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_high_noise.safetensors wget -P /home/jjj/ComfyUI/models/loras https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/loras/wan2.2_t2v_lightx2v_4steps_lora_v1.1_low_noise.safetensors wget -P /home/jjj/ComfyUI/models/text_encoders https://hf-mirror.com/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors wget -P /home/jjj/ComfyUI/models/diffusion_models https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors wget -P /home/jjj/ComfyUI/models/diffusion_models https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/diffusion_models/wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors wget -P /home/jjj/ComfyUI/models/vae https://hf-mirror.com/Comfy-Org/Wan_2.2_ComfyUI_Repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors等候下载。。。。。。。。。。。大概2小时我家网速慢华数的9MB/s的速度说实话很快了就怪模型太大在家也不干什么看着便宜就他划算老小区没光纤没法升级下车要换移动和电信了2、运行我们都用了默认的参数只是演示一下跑完了,[INFO] Prompt executed in 00:29:12时间够长我们打开enable_turbo_mode,现在不用搞清楚什么意思就看能不能跑与lora加速有关[INFO] Prompt executed in 190.14 seconds效果明显当然这个有第二次运行可能缓存里已经有数据有关。大家可以试试清楚缓存或重启comfyui后续说明下一篇我们将详细解释节点参数并且实验多种模型已经说明他们的区别