Files
gpu-stack-bootstrap/06-comfyui-image.md
T
hwd32 58fdb03c30 docs: 初始装机指南
包含 6 章步骤 + 排错手册:
- 01 硬件与基础检查
- 02 Windows 装 Studio 驱动
- 03 WSL2 + CUDA 12.8
- 04 Ollama 装 LLM
- 05 PyTorch 验证(待补,依赖实机跑通)
- 06 ComfyUI 装出图(待补,依赖实机跑通)
- 90 排错手册
2026-06-14 12:08:38 +08:00

121 lines
2.8 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# 06 ComfyUI 装出图
## 目标
装 ComfyUI + 拉 Flux.1-dev fp8 模型 + 出第一张图。
## 拉 ComfyUI 代码
**WSL2 终端**
```bash
cd /mnt/d/img-code
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
```
## 建独立 venv(和 LLM 那个分开)
```bash
uv venv --python 3.11 .venv
source .venv/bin/activate
```
## 装依赖
```bash
uv pip install -r requirements.txt
uv pip install torch torchvision torchaudio \
--index-url https://download.pytorch.org/whl/cu128 \
--extra-index-url https://pypi.tuna.tsinghua.edu.cn/simple
```
## 让 ComfyUI 用 D 盘模型
**WSL2 终端**(在 ComfyUI 目录下):
```bash
cat > extra_model_paths.yaml << 'EOF'
comfyui_models:
base_path: /mnt/d/ComfyUI-models
checkpoints: checkpoints
vae: vae
loras: loras
controlnet: controlnet
EOF
```
**这样配置后**
- 你把模型放 `/mnt/d/ComfyUI-models/checkpoints/`
- ComfyUI 自动找到(不用改代码)
## 验证 PyTorch(同 05 章)
```bash
python -c "import torch; print(torch.cuda.is_available(), torch.cuda.get_device_name(0))"
```
## 拉 Flux.1-dev fp8 模型(16G 显存能跑)
**WSL2 终端**
```bash
source .venv/bin/activate
uv pip install huggingface_hub
python -c "
from huggingface_hub import hf_hub_download
import os
os.makedirs('/mnt/d/ComfyUI-models/checkpoints', exist_ok=True)
os.makedirs('/mnt/d/ComfyUI-models/vae', exist_ok=True)
hf_hub_download(repo_id='Comfy-Org/Flux.1-FP8', filename='flux1-dev-fp8.safetensors', local_dir='/mnt/d/ComfyUI-models/checkpoints/')
hf_hub_download(repo_id='Comfy-Org/Flux.1-FP8', filename='ae.safetensors', local_dir='/mnt/d/ComfyUI-models/vae/')
print('下载完成')
"
```
**等 10-20 分钟**(约 12 GB)。
## 启动 ComfyUI
```bash
python main.py --listen 0.0.0.0
```
**看到**
```
To see the GUI go to: http://0.0.0.0:8188
```
**Windows 浏览器打开**http://localhost:8188
## 出第一张图
1. 浏览器打开 `http://localhost:8188`
2. 左键双击空白处 → 搜 "Load Checkpoint" → 选 `flux1-dev-fp8.safetensors`
3. 加 "CLIP Text Encode" 节点(输入正面提示词)
4. 加 "Empty Latent Image" 节点(设置 1024x1024
5. 加 "KSampler" 节点
6. 加 "VAE Decode" + "Save Image" 节点
7. 点 "Queue Prompt" → 等 30-60 秒
**第一次出图会编译 kernel,比较慢**(2-3 分钟),之后每张 5-10 秒。
## 模型推荐
| 模型 | 显存 | 速度 | 质量 |
|------|------|------|------|
| Flux.1-dev fp8 | 12G | 中 | ⭐⭐⭐⭐⭐ |
| SDXL | 7G | 快 | ⭐⭐⭐⭐ |
| SD 1.5 | 4G | 很快 | ⭐⭐⭐ |
**16G 显存**
- Flux.1-dev fp8 刚好
- SDXL 舒服
- SD 1.5 跑得飞快
## 关掉 ComfyUI
回到 WSL2 终端,按 `Ctrl+C`
## 下一步
✅ 通过后 → 装更多模型 / 装 ComfyUI-Manager(一键装插件)/ 学 workflow