comfyui-controlnet

ComfyUI ControlNet

使用 ControlNet 控制图像生成：姿态、边缘、深度等。

---

预注入变量

run_python 已注入：submit_workflow, folder_paths, save_preview

---

ControlNet 类型

类型	预处理器	用途	适用场景
Canny	CannyEdgePreprocessor	边缘检测	保持轮廓结构
OpenPose	OpenposePreprocessor	人体姿态	人物动作控制
Depth	DepthPreprocessor	深度图	空间关系控制
Lineart	LineartPreprocessor	线稿提取	动漫线稿上色
MLSD	M-LSDPreprocessor	直线检测	建筑/室内设计
Scribble	ScribblePreprocessor	涂鸦识别	草图生成
Seg	SemanticSegmentor	语义分割	场景布局控制
Normal	NormalMapPreprocessor	法线贴图	3D 表面细节

---

使用方法

方法 1: 使用便捷函数

from comfyui_controlnet import build_controlnet_workflow

wf = build_controlnet_workflow(
    control_type="canny",  # 或 "openpose", "depth", "lineart"
    image_path="reference.png",
    prompt="masterpiece, best quality, a girl standing",
    negative_prompt="low quality",
    checkpoint="sdxl_base.safetensors",
    controlnet_model="controlnet-canny-sdxl.safetensors",
    control_strength=0.8,  # ControlNet 强度
    width=1024,
    height=1024,
    seed=42
)

result = submit_workflow(wf)
if result.get("images"):
    save_preview(result["images"][0])

方法 2: 分步构建 Canny

import random

wf = {}

# 1. 加载模型
wf["1"] = {"class_type": "CheckpointLoaderSimple", "inputs": {"ckpt_name": "sdxl.safetensors"}}

# 2. 加载 ControlNet 模型
wf["2"] = {"class_type": "ControlNetLoader", "inputs": {"control_net_name": "control_v11p_sd15_canny.pth"}}

# 3. 加载参考图
wf["3"] = {"class_type": "LoadImage", "inputs": {"image": "pose_reference.png"}}

# 4. Canny 预处理
wf["4"] = {
    "class_type": "CannyEdgePreprocessor",
    "inputs": {
        "image": ["3", 0],
        "low_threshold": 100,
        "high_threshold": 200
    }
}

# 5. 编码提示词
wf["5"] = {"class_type": "CLIPTextEncode", "inputs": {"text": "masterpiece, best quality", "clip": ["1", 1]}}
wf["6"] = {"class_type": "CLIPTextEncode", "inputs": {"text": "low quality", "clip": ["1", 1]}}

# 6. 应用 ControlNet
wf["7"] = {
    "class_type": "ControlNetApply",
    "inputs": {
        "conditioning": ["5", 0],  # 正面提示词
        "control_net": ["2", 0],   # ControlNet 模型
        "image": ["4", 0],         # 预处理后的图像
        "strength": 0.8            # ControlNet 强度
    }
}

# 7. 生成
wf["8"] = {"class_type": "EmptyLatentImage", "inputs": {"width": 1024, "height": 1024, "batch_size": 1}}
wf["9"] = {
    "class_type": "KSampler",
    "inputs": {
        "model": ["1", 0],
        "positive": ["7", 0],  # 使用 ControlNet 后的 conditioning
        "negative": ["6", 0],
        "latent_image": ["8", 0],
        "seed": random.randint(0, 2**32),
        "steps": 20,
        "cfg": 7.0,
        "sampler_name": "euler",
        "scheduler": "normal",
        "denoise": 1.0
    }
}

# 8. 解码保存
wf["10"] = {"class_type": "VAEDecode", "inputs": {"samples": ["9", 0], "vae": ["1", 2]}}
wf["11"] = {"class_type": "SaveImage", "inputs": {"images": ["10", 0], "filename_prefix": "controlnet"}}

result = submit_workflow(wf)