【本地AI部署】comfyUI + zImageTurbo

首先声明，这是我第一次折腾这些，不是很懂，一些说法表达可能不准确，因此仅供参考。

硬件准备

电脑，我的是windows11， RTX3060 6G，配置不高，因此一些大模型就没法考虑了
关键是我电脑存储空间也不太够……

环境准备

CUDA：https://developer.nvidia.com/cuda-toolkit

下载安装。https://www.cnblogs.com/zwj/p/20220214cnda.html
（安装了cuda就行）

Anconda。不用anconda本地就装个python也行（3.10），如果想用anconda但是没安装的话，可以看：https://www.cnblogs.com/zwj/p/Anaconda-install.html
可以百度下怎么安装。

nvidia-smi

ComfyUI

参考资料：https://comfyui-wiki.com/zh/install/install-comfyui/install-comfyui-on-windows

（官方能直接下载windows exe：https://www.comfy.org/zh-cn/download）

我是看的这个博客：https://comfyui.org/zh/comfyui-windows-conda-venv

conda创建环境，下载comfyUI，安装依赖

conda create -n comfyui_env python=3.10
conda activate comfyui_env
git clone https://github.com/comfyanonymous/ComfyUI.git C:\ComfyUI
cd C:\ComfyUI
pip install -r requirements.txt

注意，不想安装C记得改，安装C请确保至少有20GB空间

如果遇到PyTorch不兼容问题，删除了重新安装

pip uninstall torch torchvision torchaudio

https://pytorch.org/get-started/locally/
这个浏览器里面，选择适合你的情况，然后复制最后的命令

安装完可以验证一下：

python -c "import torch; print(torch.cuda.is_available())"

模型

models\text_encoders：qwen_3_4b.safetensors
下载地址：https://huggingface.co/Comfy-Org/z_image_turbo/blob/main/split_files/text_encoders/qwen_3_4b.safetensors

models\vae：ae.safetensors
下载地址：https://huggingface.co/Comfy-Org/z_image_turbo/blob/main/split_files/vae/ae.safetensors

models\checkpoints：zImageTurboQuantized_fp8ScaledE4m3fnKJ.safetensors
下载地址：
https://civitai.com/models/2169712/z-image-turbo-quantized-for-low-vram?modelVersionId=2443345

放到对应文件夹，然后运行

python main.py

打开 http://127.0.0.1:8188

将工作流拖进去(工作流和comfyui-api-bridge放在了一起的，后面有下载地址，如果下载地址失效了，可以自己创建一个.json文件把下面代码复制进去)

{
  "2": {
    "inputs": {
      "text": "a beautiful landscape, high quality, 8k",
      "speak_and_recognation": {
        "__value__": [
          false,
          true
        ]
      },
      "clip": [
        "16",
        0
      ]
    },
    "class_type": "CLIPTextEncode",
    "_meta": {
      "title": "正向"
    }
  },
  "4": {
    "inputs": {
      "seed": 1065951732236213,
      "steps": 8,
      "cfg": 1,
      "sampler_name": "euler",
      "scheduler": "simple",
      "denoise": 1,
      "model": [
        "15",
        0
      ],
      "positive": [
        "2",
        0
      ],
      "negative": [
        "9",
        0
      ],
      "latent_image": [
        "5",
        0
      ]
    },
    "class_type": "KSampler",
    "_meta": {
      "title": "K采样器"
    }
  },
  "5": {
    "inputs": {
      "width": 768,
      "height": 768,
      "batch_size": 1
    },
    "class_type": "EmptyLatentImage",
    "_meta": {
      "title": "空Latent图像"
    }
  },
  "6": {
    "inputs": {
      "vae_name": "ae.safetensors"
    },
    "class_type": "VAELoader",
    "_meta": {
      "title": "加载VAE"
    }
  },
  "7": {
    "inputs": {
      "samples": [
        "4",
        0
      ],
      "vae": [
        "6",
        0
      ]
    },
    "class_type": "VAEDecode",
    "_meta": {
      "title": "VAE解码"
    }
  },
  "8": {
    "inputs": {
      "filename_prefix": "ComfyUI",
      "images": [
        "7",
        0
      ]
    },
    "class_type": "SaveImage",
    "_meta": {
      "title": "保存图像"
    }
  },
  "9": {
    "inputs": {
      "text": "blurry, ugly, bad, lowres, jpeg artifacts, watermark, distorted, noisy, artifact, glitch, oversaturation, neon tones, harsh contrast or glow, color cast, pixelated, blocky",
      "speak_and_recognation": {
        "__value__": [
          false,
          true
        ]
      },
      "clip": [
        "16",
        0
      ]
    },
    "class_type": "CLIPTextEncode",
    "_meta": {
      "title": "反向"
    }
  },
  "15": {
    "inputs": {
      "ckpt_name": "zImageTurboQuantized_fp8ScaledE4m3fnKJ.safetensors"
    },
    "class_type": "CheckpointLoaderSimple",
    "_meta": {
      "title": "Checkpoint加载器（简易）"
    }
  },
  "16": {
    "inputs": {
      "clip_name": "qwen_3_4b.safetensors",
      "type": "stable_diffusion",
      "device": "default"
    },
    "class_type": "CLIPLoader",
    "_meta": {
      "title": "加载CLIP"
    }
  }
}