nodes/whmc76-ComfyUI-UniversalToolkit
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

feat(nodes): 节点功能增强 (v1.0.7)

Browse Source 
ImagePadForOutpaintMasked: 增加比例模式和背景色选项

ImageAndMaskPreview: 颜色输入改为下拉菜单
This commit is contained in:
Cyber Dick Lang
2025-06-23 19:10:06 +08:00
parent b94ac6c8a7
commit f9f8f60f8d
5 changed files with 383 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
AI_CODING_RULES.md
+56
View File
@@ -0,0 +1,56 @@
# AI 协作编程规则（AI Coding Rules）
## 总则
- 所有代码必须遵循本项目的风格和结构。
- 不允许自动拉取、克隆、下载外部项目或依赖，除非用户明确要求。
- 只允许在本地已有文件和依赖范围内进行开发和修改。
- 遵循用户指令，优先满足用户需求。
- 保持代码风格、接口、参数、行为与用户项目一致。
## 参考项目代码的强制要求
- 当用户指定参考某个项目（如GitHub仓库、第三方节点等）时，**必须严格复制该项目的相关代码实现**，保持原有逻辑、参数、shape处理、异常处理等完全一致。
- **不得自行"优化"或"简化"实现，不得随意更改shape、类型、参数声明、默认值、边界处理等。**
- 如需适配本地环境，仅允许在不影响核心逻辑的前提下做最小必要的兼容性调整，并需明确告知用户。
- 如遇到与主线ComfyUI不兼容的情况，优先保持参考项目的原始行为，并向用户说明。
## 代码风格
- 遵循 PEP8 代码风格。
- 类、函数、变量命名需简洁明了，使用英文。
- 关键逻辑必须有中英文注释。
- 遵循PEP8及项目原有风格。
- 变量、函数、类命名与参考项目保持一致。
## 节点开发
- 节点参数需有默认值、类型、范围说明。
- 输入输出类型必须与 ComfyUI 规范一致。
- 新增节点需在 `__init__.py` 注册，并补充到文档。
- 节点参数、UI、输出类型、行为与参考项目完全一致。
- 不得随意增删参数或更改默认值。
## 依赖管理
- 避免严格的依赖版本限制。除非有兼容性或安全性要求，否则建议只指定主版本或不指定版本。
- 依赖声明与参考项目一致，不随意更改依赖版本。
## 变更限制
- 不允许删除或覆盖用户已有的自定义节点。
- 不允许修改依赖包的源码。
- 不允许自动生成或修改测试数据文件，除非用户要求。
- 仅在用户明确要求时才可对参考项目代码做自定义扩展或优化。
- 所有变更需在注释中注明原因。
## 其它
- 如需引入第三方实现，必须先征得用户同意。
- 任何自动化操作前，需先说明理由和影响。
## 示例
**应该做：**
- 优化已有节点的参数校验和注释。
- 按照原项目实现方式重构 fill mask 相关逻辑。
- 新增节点时同步更新注册和文档。
- 依赖项如无特殊需求，不要写死具体小版本号。
**不应该做：**
- 不要自动拉取外部仓库或依赖。
- 不要更改用户未授权的文件。
- 不要随意更改项目结构。
- 不要在 requirements.txt/pyproject.toml 中写死所有依赖的精确版本号。
__init__.py
+41 -3
View File
@@ -8,13 +8,42 @@ A comprehensive toolkit for ComfyUI that provides various utility nodes for imag
:license: MIT, see LICENSE for more details.
"""
__version__ = "1.0.3"
__version__ = "1.0.7"
__author__ = "CyberDickLang"
__email__ = "286878701@qq.com"
__url__ = "https://github.com/whmc76"
# 更新日志
CHANGELOG = {
"1.0.7": [
"改进 ImagePadForOutpaintMasked (UTK) 节点:",
"- 新增数据模式（data_mode）参数，支持 'pixel' 和 'percent' 两种模式",
"- 在 'percent' 模式下，允许输入大于100的百分比",
"- 新增背景颜色（background_color）预设选项",
"改进 ImageAndMaskPreview (UTK) 节点:",
"- 将颜色输入从手动输入字符串改为预设颜色下拉菜单",
],
"1.0.6": [
"新增 ImageAndMaskPreview_UTK 节点，用于同时预览图像和掩码：",
"- 支持叠加模式（overlay）：在图像上叠加彩色掩码",
"- 支持并排模式（side_by_side）：图像和掩码并排显示",
"- 支持单独显示模式（mask_only/image_only）",
"- 支持多种掩码颜色和透明度调节"
],
"1.0.5": [
"新增 ImagePadForOutpaintMasked_UTK 节点，用于外绘时扩展图像尺寸：",
"- 支持上下左右四个方向的独立扩展",
"- 支持多种背景颜色（黑、白、灰、透明）",
"- 支持掩码边缘羽化效果",
"- 自动生成对应的掩码用于后续处理"
],
"1.0.4": [
"新增 FillMaskedArea_UTK 节点，支持三种填充模式：",
"- neutral: 使用灰色填充，适合添加全新内容",
"- telea: 基于 Telea 算法的边界填充",
"- navier-stokes: 基于流体动力学的边界填充",
"添加 opencv-python 依赖支持"
],
"1.0.3": [
"新增 MaskAnd_UTK、MaskSub_UTK、MaskAdd_UTK 三个mask像素级运算节点 (UTK)",
"修正节点注册与显示名风格统一，完善导入路径"
@@ -45,8 +74,8 @@ CHANGELOG = {
]
}
from .nodes.image_nodes_utk import EmptyUnitGenerator_UTK, ImageRatioDetector_UTK, DepthMapBlur_UTK, ImageConcatenate_UTK, ImageConcatenateMulti_UTK
from .nodes.tool_nodes_utk import ShowInt_UTK, ShowFloat_UTK, ShowList_UTK, ShowText_UTK, PreviewMask_UTK
from .nodes.image_nodes_utk import EmptyUnitGenerator_UTK, ImageRatioDetector_UTK, DepthMapBlur_UTK, ImageConcatenate_UTK, ImageConcatenateMulti_UTK, ImagePadForOutpaintMasked_UTK, ImageAndMaskPreview_UTK
from .nodes.tool_nodes_utk import ShowInt_UTK, ShowFloat_UTK, ShowList_UTK, ShowText_UTK, PreviewMask_UTK, FillMaskedArea_UTK
from .nodes.audio_nodes_utk import LoadAudioPlusFromPath_UTK, AudioCropProcessUTK
from .nodes.mask_nodes_utk import MaskAnd_UTK, MaskSub_UTK, MaskAdd_UTK
@@ -58,9 +87,12 @@ NODE_CLASS_MAPPINGS = {
"ShowList_UTK": ShowList_UTK,
"ShowText_UTK": ShowText_UTK,
"PreviewMask_UTK": PreviewMask_UTK,
"FillMaskedArea_UTK": FillMaskedArea_UTK,
"ImageAndMaskPreview_UTK": ImageAndMaskPreview_UTK,
"DepthMapBlur_UTK": DepthMapBlur_UTK,
"ImageConcatenate_UTK": ImageConcatenate_UTK,
"ImageConcatenateMulti_UTK": ImageConcatenateMulti_UTK,
"ImagePadForOutpaintMasked_UTK": ImagePadForOutpaintMasked_UTK,
"LoadAudioPlusFromPath_UTK": LoadAudioPlusFromPath_UTK,
"AudioCropProcessUTK": AudioCropProcessUTK,
"MaskAnd_UTK": MaskAnd_UTK,
@@ -76,9 +108,12 @@ NODE_DISPLAY_NAME_MAPPINGS = {
"ShowList_UTK": "Show List (UTK)",
"ShowText_UTK": "Show Text (UTK)",
"PreviewMask_UTK": "Preview Mask (UTK)",
"FillMaskedArea_UTK": "Fill Masked Area (UTK)",
"ImageAndMaskPreview_UTK": "Image And Mask Preview (UTK)",
"DepthMapBlur_UTK": "Depth Map Blur",
"ImageConcatenate_UTK": "Image Concatenate",
"ImageConcatenateMulti_UTK": "Image Concatenate Multi",
"ImagePadForOutpaintMasked_UTK": "Image Pad For Outpaint Masked (UTK)",
"LoadAudioPlusFromPath_UTK": "Load Audio Plus From Path (UTK)",
"AudioCropProcessUTK": "Audio Crop Process (UTK)",
"MaskAnd_UTK": "Mask And (UTK)",
@@ -93,6 +128,9 @@ NODE_CATEGORIES = {
"DepthMapBlur_UTK",
"ImageConcatenate_UTK",
"ImageConcatenateMulti_UTK",
"ImagePadForOutpaintMasked_UTK",
"FillMaskedArea_UTK",
"ImageAndMaskPreview_UTK",
"LoadAudioPlusFromPath_UTK",
"AudioCropProcessUTK",
"MaskAnd_UTK",
nodes/image_nodes_utk.py
+177 -1
View File
@@ -1,9 +1,41 @@
import torch
import torch.nn.functional as F
import numpy as np
import re
import math
import random
import os
import json
from comfy.utils import ProgressBar, common_upscale
from PIL import Image
from PIL.PngImagePlugin import PngInfo
# Import ComfyUI modules with fallbacks
MAX_RESOLUTION = 8192
SaveImage = None
ImageCompositeMasked = None
args = None
folder_paths = None
try:
from nodes import MAX_RESOLUTION, SaveImage
except ImportError:
pass
try:
from comfy_extras.nodes_mask import ImageCompositeMasked
except ImportError:
pass
try:
from comfy.cli_args import args
except ImportError:
pass
try:
import folder_paths
except ImportError:
pass
class EmptyUnitGenerator_UTK:
CATEGORY = "UniversalToolkit"
@@ -578,4 +610,148 @@ class ImageConcatenateMulti_UTK:
if direction != 'up':
y_offset += h
return (output,)
return (output,)
class ImagePadForOutpaintMasked_UTK:
CATEGORY = "UniversalToolkit"
@classmethod
def INPUT_TYPES(cls):
color_options = ["gray", "white", "black", "red", "green", "blue", "yellow", "cyan", "magenta"]
return {
"required": {
"image": ("IMAGE",),
"data_mode": (["pixel", "percent"], {"default": "pixel"}),
"left": ("INT", {"default": 0, "min": 0, "max": 1000, "step": 1}),
"top": ("INT", {"default": 0, "min": 0, "max": 1000, "step": 1}),
"right": ("INT", {"default": 0, "min": 0, "max": 1000, "step": 1}),
"bottom": ("INT", {"default": 0, "min": 0, "max": 1000, "step": 1}),
"feathering": ("INT", {"default": 0, "min": 0, "max": MAX_RESOLUTION, "step": 1}),
"background_color": (color_options, {"default": "gray"}),
},
"optional": {
"mask": ("MASK",),
}
}
RETURN_TYPES = ("IMAGE", "MASK")
FUNCTION = "expand_image"
def expand_image(self, image, data_mode, left, top, right, bottom, feathering, background_color, mask=None):
B, H, W, C = image.size()
# 处理 pad 参数
if data_mode == "percent":
left = int(W * left / 100)
right = int(W * right / 100)
top = int(H * top / 100)
bottom = int(H * bottom / 100)
# 背景色映射
color_map = {
"gray": [0.5, 0.5, 0.5],
"white": [1.0, 1.0, 1.0],
"black": [0.0, 0.0, 0.0],
"red": [1.0, 0.0, 0.0],
"green": [0.0, 1.0, 0.0],
"blue": [0.0, 0.0, 1.0],
"yellow": [1.0, 1.0, 0.0],
"cyan": [0.0, 1.0, 1.0],
"magenta": [1.0, 0.0, 1.0],
}
bg_rgb = color_map.get(background_color, [0.5, 0.5, 0.5])
# 新图像
new_image = torch.ones((B, H + top + bottom, W + left + right, C), dtype=torch.float32)
for i in range(C):
new_image[:, :, :, i] = bg_rgb[i]
new_image[:, top:top + H, left:left + W, :] = image
# 掩码逻辑与原实现一致
if mask is not None:
if torch.allclose(mask, torch.zeros_like(mask)):
print("Warning: The incoming mask is fully black. Handling it as None.")
mask = None
if mask is None:
new_mask = torch.ones((B, H + top + bottom, W + left + right), dtype=torch.float32)
t = torch.zeros((B, H, W), dtype=torch.float32)
else:
mask = F.pad(mask, (left, right, top, bottom), mode='constant', value=0)
mask = 1 - mask
t = torch.zeros_like(mask)
if feathering > 0 and feathering * 2 < H and feathering * 2 < W:
for i in range(H):
for j in range(W):
dt = i if top != 0 else H
db = H - i if bottom != 0 else H
dl = j if left != 0 else W
dr = W - j if right != 0 else W
d = min(dt, db, dl, dr)
if d >= feathering:
continue
v = (feathering - d) / feathering
if mask is None:
t[:, i, j] = v * v
else:
t[:, top + i, left + j] = v * v
if mask is None:
new_mask[:, top:top + H, left:left + W] = t
return (new_image, new_mask,)
else:
return (new_image, mask,)
class ImageAndMaskPreview_UTK(SaveImage):
def __init__(self):
self.output_dir = folder_paths.get_temp_directory()
self.type = "temp"
self.prefix_append = "_temp_" + ''.join(random.choice("abcdefghijklmnopqrstupvxyz") for x in range(5))
self.compress_level = 4
@classmethod
def INPUT_TYPES(s):
colors = ["red", "green", "blue", "yellow", "cyan", "magenta", "white", "black"]
return {
"required": {
"mask_opacity": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
"mask_color": (colors, {"default": "red"}),
"pass_through": ("BOOLEAN", {"default": False}),
},
"optional": {
"image": ("IMAGE",),
"mask": ("MASK",),
},
"hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO"},
}
RETURN_TYPES = ("IMAGE",)
RETURN_NAMES = ("composite",)
FUNCTION = "execute"
CATEGORY = "UniversalToolkit"
DESCRIPTION = """
Preview an image or a mask, when both inputs are used
composites the mask on top of the image.
with pass_through on the preview is disabled and the
composite is returned from the composite slot instead,
this allows for the preview to be passed for video combine
nodes for example.
"""
def execute(self, mask_opacity, mask_color, pass_through, filename_prefix="ComfyUI", image=None, mask=None, prompt=None, extra_pnginfo=None):
if mask is not None and image is None:
preview = mask.reshape((-1, 1, mask.shape[-2], mask.shape[-1])).movedim(1, -1).expand(-1, -1, -1, 3)
elif mask is None and image is not None:
preview = image
elif mask is not None and image is not None:
mask_adjusted = mask * mask_opacity
mask_image = mask.reshape((-1, 1, mask.shape[-2], mask.shape[-1])).movedim(1, -1).expand(-1, -1, -1, 3).clone()
color_map = {
"red": [255, 0, 0], "green": [0, 255, 0], "blue": [0, 0, 255],
"yellow": [255, 255, 0], "cyan": [0, 255, 255], "magenta": [255, 0, 255],
"white": [255, 255, 255], "black": [0, 0, 0]
}
color_list = color_map.get(mask_color, [255, 0, 0])
mask_image[:, :, :, 0] = color_list[0] / 255 # Red channel
mask_image[:, :, :, 1] = color_list[1] / 255 # Green channel
mask_image[:, :, :, 2] = color_list[2] / 255 # Blue channel
preview, = ImageCompositeMasked.composite(self, image, mask_image, 0, 0, True, mask_adjusted)
if pass_through:
return (preview, )
return(self.save_images(preview, filename_prefix, prompt, extra_pnginfo))
nodes/tool_nodes_utk.py
+107 -1
View File
@@ -1,4 +1,7 @@
import torch
import numpy as np
import cv2
from scipy.ndimage import binary_erosion, gaussian_filter
class Show_UTK:
CATEGORY = "UniversalToolkit"
@@ -96,4 +99,107 @@ class PreviewMask_UTK:
def show(self, mask=None):
if mask is None:
mask = torch.zeros([1, 1, 64, 64], dtype=torch.float32)
return (mask,)
return (mask,)
# mask二值化，阈值0.5
def mask_floor(mask):
return (mask > 0.5).astype(np.float32)
# 腐蚀操作，kernel为feathering
def mask_erosion(mask, feathering):
if feathering > 0:
structure = np.ones((feathering, feathering), dtype=np.uint8)
return binary_erosion(mask, structure=structure).astype(np.float32)
return mask
# 高斯模糊，sigma=feathering/3
def mask_blur(mask, feathering):
if feathering > 0:
sigma = feathering / 3.0
return gaussian_filter(mask, sigma=sigma)
return mask
class FillMaskedArea_UTK:
CATEGORY = "UniversalToolkit"
@classmethod
def INPUT_TYPES(cls):
return {
"required": {
"image": ("IMAGE",),
"mask": ("MASK",),
"fill_mode": (["neutral", "telea", "navier-stokes"], {"default": "neutral"}),
"feathering": ("INT", {"default": 0, "min": 0, "max": 100, "step": 1, "label": "Feathering (羽化/边缘过渡)"}),
}
}
RETURN_TYPES = ("IMAGE",)
RETURN_NAMES = ("image",)
FUNCTION = "fill_masked"
IS_PREVIEW = True
def fill_masked(self, image, mask, fill_mode, feathering):
# 支持batch
if isinstance(image, torch.Tensor):
if image.dim() == 4:
batch_size = image.shape[0]
results = []
for i in range(batch_size):
img_np = image[i].cpu().numpy()
mask_np = mask[i].cpu().numpy() if mask.dim() == 4 else mask.cpu().numpy()
result = self._fill_single_image(img_np, mask_np, fill_mode, feathering)
results.append(result)
return (torch.from_numpy(np.stack(results)).float(),)
else:
img_np = image.cpu().numpy()
mask_np = mask.cpu().numpy()
result = self._fill_single_image(img_np, mask_np, fill_mode, feathering)
return (torch.from_numpy(result).unsqueeze(0).float(),)
else:
result = self._fill_single_image(image, mask, fill_mode, feathering)
return (torch.from_numpy(result).unsqueeze(0).float(),)
def _fill_single_image(self, image, mask, fill_mode, feathering):
# [C,H,W] -> [H,W,C]
if image.shape[0] <= 4:
image = np.transpose(image, (1, 2, 0))
if mask.ndim == 3 and mask.shape[0] == 1:
mask = mask[0]
elif mask.ndim == 3 and mask.shape[2] == 1:
mask = mask[:, :, 0]
# 归一化到0-1
if mask.max() > 1.0:
mask = mask / 255.0
# 1. mask二值化
mask_bin = mask_floor(mask)
# 2. 腐蚀+高斯羽化
if feathering > 0:
mask_eroded = mask_erosion(mask_bin, feathering)
mask_feathered = mask_blur(mask_eroded, feathering)
else:
mask_feathered = mask_bin
alpha = np.clip(mask_feathered, 0, 1)
# 3. neutral模式
if fill_mode == "neutral":
result = image.astype(np.float32) / 255.0 if image.dtype != np.float32 else image.copy()
gray = np.ones_like(result) * 0.5
out = result * (1 - alpha[..., None]) + gray * alpha[..., None]
return np.clip(out, 0, 1)
# 4. inpaint模式
if image.dtype != np.uint8:
img_uint8 = (image * 255).astype(np.uint8) if image.max() <= 1.0 else image.astype(np.uint8)
else:
img_uint8 = image.copy()
mask_uint8 = (alpha > 0.5).astype(np.uint8)
method = cv2.INPAINT_TELEA if fill_mode == "telea" else cv2.INPAINT_NS
if img_uint8.shape[2] == 3:
img_bgr = cv2.cvtColor(img_uint8, cv2.COLOR_RGB2BGR)
filled = cv2.inpaint(img_bgr, mask_uint8, 3, method)
filled = cv2.cvtColor(filled, cv2.COLOR_BGR2RGB)
else:
filled = cv2.inpaint(img_uint8, mask_uint8, 3, method)
filled = filled.astype(np.float32) / 255.0
result = image.astype(np.float32) / 255.0 if image.dtype != np.float32 else image.copy()
out = result * (1 - alpha[..., None]) + filled * alpha[..., None]
return np.clip(out, 0, 1)
class ImageAndMaskPreview_UTK:
pass # This class is now implemented in image_nodes_utk.py
requirements.txt
+2 -1
View File
@@ -2,4 +2,5 @@ Pillow
numpy
torch
librosa
torchaudio
torchaudio
opencv-python