Add files via upload
This commit is contained in:
@@ -1,131 +1,128 @@
|
||||
# ComfyUI-Advanced-Tile-Processing
|
||||
|
||||
**ComfyUI-Advanced-Tile-Processing** 是一套专为 ComfyUI 设计的高级自定义节点插件。它旨在解决超高分辨率图像生成和处理中的显存瓶颈(VRAM OOM),通过智能分块(Tiling)与基于加权累积的无缝合并(Merging)技术,让您能够在普通消费级显卡上稳定处理 4K、8K 甚至更高规格的图像。
|
||||
|
||||
## 🌟 核心特性
|
||||
|
||||
- **智能维度感知 (Smart Dimension Awareness)**:自动计算最优切片坐标。支持“边缘回退”策略,确保分块完全覆盖原始图像,无需引入黑边(Padding)从而避免模型产生伪影。
|
||||
- **无缝加权融合 (Soft Blending)**:内置多种混合模式(Gaussian, Cosine, Linear)。通过多层重叠区域的加权平均,最大化消除物理拼接产生的“网格接缝”。
|
||||
- **元数据驱动流 (Metadata-Driven Workflow)**:Splitter 生成唯一的 `TILE_CONFIG` 对象,包含原始尺寸、缩放因子及坐标。Merger 自动读取配置一键还原,用户无需手动对齐参数。
|
||||
- **深度兼容性**:完美适配 ComfyUI 的列表执行机制。支持将分块作为 `BATCH` 提交以获得最高推理速度,或作为 `LIST` 提交以支持循环节点处理。显存不足的情况下推荐 `LIST` 替代 `BATCH` 。
|
||||
- **高性能架构**:完全基于 PyTorch 张量运算,GPU 加速,保证处理处理大图时性能在线。
|
||||
- **动态尺寸支持 (Tiled Upscale)**:支持动态尺寸调整(Tiled Upscale),在分块后进行放大处理。Merger 能够自动检测片段尺寸变化并动态调整画布大小。
|
||||
|
||||
## 🛠 安装说明
|
||||
|
||||
1. **环境要求**:ComfyUI 0.4.0+,Python 3.10+,PyTorch 2.0+。
|
||||
|
||||
2. 进入您的 ComfyUI 自定义节点目录:
|
||||
|
||||
```
|
||||
cd ComfyUI/custom_nodes/
|
||||
```
|
||||
|
||||
3. 克隆本仓库:
|
||||
```
|
||||
git clone [https://github.com/QL-boy/ComfyUI-Advanced-Tile-Processing.git](https://github.com/QL-boy/ComfyUI-Advanced-Tile-Processing.git)
|
||||
```
|
||||
|
||||
4. 重启 ComfyUI。
|
||||
|
||||
|
||||
## 🧩 节点详解
|
||||
|
||||
### 1. 🔧 Advanced Tile Splitter (分块器)
|
||||

|
||||
将输入的大图或 Latent 空间分割成重叠的小块。
|
||||
|
||||
- **输入端口**:
|
||||
|
||||
- `image`: (可选) 原始图像。
|
||||
- `latent`: (可选) 潜在空间数据。
|
||||
|
||||
- **核心参数**:
|
||||
|
||||
- `tile_size`: 分块分辨率(如 512, 1024)。
|
||||
- `overlap`: 两个分块之间的重叠像素。建议设为 `tile_size` 的 10% 及以上以获得最佳融合效果。
|
||||
|
||||
- **输出端口**:
|
||||
|
||||
- `tiles_image_batch`: 将所有分块合并为一个 Batch Tensor,适合高性能采样。
|
||||
- `tiles_image_list`: (List 模式) 分块图像列表,触发 ComfyUI 循环执行。
|
||||
- `tiles_latent_batch / list`: 对应 Latent 空间的输出。
|
||||
- `tile_config`: 核心配置文件(必需连接至 Merger),包含合并所需的所有元数据。
|
||||
|
||||
### 2. 🔧 Advanced Tile Merger (合并器)
|
||||

|
||||
根据分块配置将处理后的片段无缝还原。
|
||||
|
||||
- **输入端口**:
|
||||
|
||||
- `tile_config`: 由 Splitter 输出的配置对象。
|
||||
- `processed_tiles_image`: (可选) 处理后的图像片段。
|
||||
- `processed_tiles_latent`: (可选) 处理后的 Latent 片段。
|
||||
|
||||
- **核心参数**:
|
||||
|
||||
- `blend_mode`:
|
||||
- `gaussian` (默认): 中心权重高,边缘平滑衰减,融合效果最自然。
|
||||
- `cosine`: 经典的三角函数衰减,过渡区域较宽。
|
||||
- `linear`: 线性金字塔混合。
|
||||
- `none`: 硬边缘拼接(会有明显接缝)。
|
||||
- `feather_percent`: 羽化比例 (0-50%),控制权重衰减的起始位置,值越大,边缘融合区域越宽。
|
||||
|
||||
- 内部机制:
|
||||
|
||||
节点会自动解包 `LIST` 输入。如果您的上游是循环节点,请确保所有分块均已处理完成再输入 Merger。
|
||||
|
||||
## 📖 核心算法原理:加权融合(Soft Blending / 累积缓冲加权平均法)
|
||||
|
||||
为消除传统“剪切-粘贴”方式产生的接缝与亮度突变,本插件采用**累积缓冲加权平均法(Accumulation Buffer Weighted Averaging)**,实现多层像素在重叠区域的平滑融合。其核心流程如下:
|
||||
|
||||
1. **掩码生成(Weight Mask Generation)**
|
||||
为每个图像分块(Tile)生成一个权重掩码矩阵 $M$。在 Gaussian 模式下,中心点权重为 1.0,并向边缘按指数规律衰减至接近 0,确保融合时中心区域贡献最大、边缘逐渐过渡。
|
||||
2. **像素累加与权重累加**
|
||||
- **像素累加**:在全局画布对应位置累加当前分块加权后的像素值:
|
||||
$Canvas += Tile \times M$
|
||||
- **权重累加**:同步在权重分布图(Weight Map)中累加掩码值:
|
||||
$WeightMap += M$
|
||||
3. **归一化输出(Normalization)**
|
||||
所有分块累加完成后,对每个像素进行归一化计算,得到最终图像:
|
||||
$Pixel_{final} = \frac{Canvas}{WeightMap + \epsilon} = \frac{\sum (Tile_i \times M_i)}{\sum M_i + \epsilon}$
|
||||
其中 $\epsilon$ 是为避免除零而设置的极小常数。
|
||||
### 关键优势
|
||||
- **无缝过渡**:重叠区域由多个 Tile 按权重共同贡献,通过归一化保持亮度与对比度一致;
|
||||
- **抗伪影**:权重平滑变化避免硬边缘,实现视觉上的自然渐变;
|
||||
- **灵活性**:可通过调整权重衰减曲线适应不同分块重叠策略。
|
||||
|
||||
该方法本质上是一种**软混合(Soft Blending)**,通过在重叠区域进行多图层像素的加权平均,达成高质量、无接缝的大图合成效果。
|
||||
## 🚀 示例工作流
|
||||
### SDXL 经典分块采样高清修复复刻
|
||||

|
||||
|
||||
### Z-Image质量升级
|
||||

|
||||
|
||||
###
|
||||
|
||||
## ⚠️ 常见问题 (FAQ)
|
||||
|
||||
- **Q: 为什么合并后的图边缘还是有淡淡的印子?**
|
||||
|
||||
- A: 尝试增加 `overlap` (建议不低于 64) 并确保 `blend_mode` 设为 `gaussian`。同时,检查重绘时的 `denoise` 是否过高,过高的降噪会导致分块内容产生剧烈变化。
|
||||
|
||||
- **Q: 支持 Latent 空间合并吗?**
|
||||
|
||||
- A: 支持。但由于 VAE 编码器的特性,Latent 空间的重叠合并可能会在 Decode 后出现微小色差,通常推荐在 Image 空间执行最终合并。
|
||||
|
||||
- **Q: 为什么我连接了节点却输出了多张图?**
|
||||
|
||||
- A: 请确保您使用的是 `Advanced Tile Merger` 并正确连接了 `tile_config`。如果输出依然是列表,请检查是否在 Merger 之后又连接了不支持 List 的旧版节点。
|
||||
|
||||
- **Q: 为什么无法在 [ComfyUI Manager](https://github.com/Comfy-Org/ComfyUI-Manager) 中安装?**
|
||||
- A: 当前还未发布到自定义节点的公共集合中,已加入待办。
|
||||
|
||||
## 🤝 贡献与反馈
|
||||
|
||||
欢迎提交 Issue 或 Pull Request。如果您在使用中发现本项目可以改进的地方,请随时联系。
|
||||
|
||||
## 📜 许可说明
|
||||
|
||||
本项目基于 [Apache-2.0 license](https://github.com/QL-boy/comfyui-ps-plugin#Apache-2.0-1-ov-file) 许可证开源。
|
||||
# ComfyUI-Advanced-Tile-Processing
|
||||
|
||||
**ComfyUI-Advanced-Tile-Processing** is an advanced custom node plugin suite designed specifically for ComfyUI. It aims to solve VRAM out-of-memory (OOM) bottlenecks in ultra-high-resolution image generation and processing by employing intelligent tiling and seamless merging techniques based on weighted accumulation, enabling stable processing of 4K, 8K, and even higher resolution images on standard consumer-grade graphics cards.
|
||||
|
||||
## 🌟 Core Features
|
||||
|
||||
- **Smart Dimension Awareness**: Automatically calculates optimal slicing coordinates. Supports "edge fallback" strategy to ensure complete coverage of the original image without introducing black borders (padding), thus avoiding artifacts generated by models.
|
||||
- **Seamless Weighted Fusion**: Built-in multiple blending modes (Gaussian, Cosine, Linear). Through weighted averaging of overlapping regions across multiple layers, it maximizes the elimination of "grid seams" produced by physical stitching.
|
||||
- **Metadata-Driven Workflow**: The Splitter generates a unique `TILE_CONFIG` object containing original dimensions, scaling factors, and coordinates. The Merger automatically reads configurations for one-click restoration without manual parameter alignment.
|
||||
- **Deep Compatibility**: Perfectly compatible with ComfyUI's list execution mechanism. Supports submitting tiles as `BATCH` for maximum inference speed or as `LIST` for loop node processing. When VRAM is insufficient, `LIST` is recommended over `BATCH`.
|
||||
- **High-Performance Architecture**: Fully based on PyTorch tensor operations with GPU acceleration, ensuring performance remains optimal when handling large images.
|
||||
- **Dynamic Size Support (Tiled Upscale)**: Supports dynamic size adjustment (Tiled Upscale), allowing upscaling after tiling. The Merger can automatically detect fragment size changes and dynamically adjust canvas size.
|
||||
|
||||
## 🛠 Installation Instructions
|
||||
|
||||
1. **Environment Requirements**: ComfyUI 0.4.0+, Python 3.10+, PyTorch 2.0+.
|
||||
|
||||
2. Navigate to your ComfyUI custom nodes directory:
|
||||
|
||||
```
|
||||
cd ComfyUI/custom_nodes/
|
||||
```
|
||||
|
||||
3. Clone this repository:
|
||||
```
|
||||
git clone [https://github.com/QL-boy/ComfyUI-Advanced-Tile-Processing.git](https://github.com/QL-boy/ComfyUI-Advanced-Tile-Processing.git)
|
||||
```
|
||||
|
||||
4. Restart ComfyUI.
|
||||
|
||||
|
||||
## 🧩 Node Details
|
||||
|
||||
### 1. 🔧 Advanced Tile Splitter (Splitter)
|
||||

|
||||
Splits the input large image or latent space into overlapping smaller tiles.
|
||||
|
||||
- **Input Ports**:
|
||||
|
||||
- `image`: (optional) Original image.
|
||||
- `latent`: (optional) Latent space data.
|
||||
|
||||
- **Core Parameters**:
|
||||
|
||||
- `tile_size`: Tile resolution (e.g., 512, 1024).
|
||||
- `overlap`: Overlapping pixels between two tiles. Recommended to be 10% or more of `tile_size` for optimal blending.
|
||||
|
||||
- **Output Ports**:
|
||||
|
||||
- `tiles_image_batch`: Merges all tiles into a Batch Tensor for high-performance sampling.
|
||||
- `tiles_image_list`: (List mode) List of tile images, triggering ComfyUI loop execution.
|
||||
- `tiles_latent_batch / list`: Corresponding outputs for latent space.
|
||||
- `tile_config`: Core configuration file (must be connected to Merger), containing all metadata required for merging.
|
||||
|
||||
### 2. 🔧 Advanced Tile Merger (Merger)
|
||||

|
||||
Seamlessly reconstructs processed fragments based on tile configuration.
|
||||
|
||||
- **Input Ports**:
|
||||
|
||||
- `tile_config`: Configuration object output by Splitter.
|
||||
- `processed_tiles_image`: (optional) Processed image fragments.
|
||||
- `processed_tiles_latent`: (optional) Processed latent fragments.
|
||||
|
||||
- **Core Parameters**:
|
||||
|
||||
- `blend_mode`:
|
||||
- `gaussian` (default): Higher weight at the center, smoothly decaying towards the edges, providing the most natural blending.
|
||||
- `cosine`: Classic trigonometric decay, with a wider transition area.
|
||||
- `linear`: Linear pyramid blending.
|
||||
- `none`: Hard edge stitching (will have noticeable seams).
|
||||
- `feather_percent`: Feathering percentage (0-50%), controlling the starting position of weight decay, with higher values resulting in wider edge blending areas.
|
||||
|
||||
- Internal Mechanism:
|
||||
|
||||
The node automatically unpacks `LIST` inputs. If your upstream is a loop node, ensure all tiles have been processed before feeding them into the Merger.
|
||||
|
||||
## 📖 Core Algorithm Principle: Weighted Fusion (Soft Blending / Accumulation Buffer Weighted Averaging)
|
||||
|
||||
To eliminate seams and brightness variations caused by traditional "cut-paste" methods, this plugin employs **Accumulation Buffer Weighted Averaging**, achieving smooth fusion of multiple pixels in overlapping regions. The core process is as follows:
|
||||
|
||||
1. **Weight Mask Generation**
|
||||
Generates a weight mask matrix $M$ for each image tile. In Gaussian mode, the center point has a weight of 1.0, decaying exponentially towards the edges, ensuring that the center region contributes the most during blending while the edges gradually transition.
|
||||
2. **Pixel Accumulation and Weight Accumulation**
|
||||
- **Pixel Accumulation**: Adds the weighted pixel values of the current tile to the corresponding position on the global canvas:
|
||||
$Canvas += Tile \times M$
|
||||
- **Weight Accumulation**: Simultaneously adds the mask values to the weight map:
|
||||
$WeightMap += M$
|
||||
3. **Normalization Output**
|
||||
After all tiles have been accumulated, each pixel is normalized to obtain the final image:
|
||||
$Pixel_{final} = \frac{Canvas}{WeightMap + \epsilon} = \frac{\sum (Tile_i \times M_i)}{\sum M_i + \epsilon}$
|
||||
where $\epsilon$ is a small constant to prevent division by zero.
|
||||
### Key Advantages
|
||||
- **Seamless Transition**: Overlapping regions are contributed by multiple tiles based on weights, maintaining brightness and contrast consistency through normalization;
|
||||
- **Artifact Resistance**: Smooth weight changes avoid hard edges, achieving a natural visual gradient;
|
||||
- **Flexibility**: Can adapt to different tile overlap strategies by adjusting the weight decay curve.
|
||||
|
||||
This method essentially performs **soft blending** by averaging pixel values across multiple layers in overlapping regions, achieving high-quality, seamless large image synthesis.
|
||||
## 🚀 Example Workflow
|
||||
### SDXL Classic Tiling Sampling High-Resolution Repair and Reproduction
|
||||

|
||||
|
||||
### Z-Image Quality Upgrade
|
||||

|
||||
|
||||
###
|
||||
|
||||
## ⚠️ Common Issues (FAQ)
|
||||
|
||||
- **Q: Why are there faint marks at the edges after merging?**
|
||||
|
||||
- A: Try increasing `overlap` (recommended to be at least 64) and ensure `blend_mode` is set to `gaussian`. Additionally, check if the `denoise` value during redrawing is too high, as excessive denoising can cause significant changes in tile content.
|
||||
|
||||
- **Q: Does it support merging latent space?**
|
||||
|
||||
- A: Yes. However, due to the characteristics of the VAE encoder, overlapping merges in latent space may result in minor color differences upon decoding. It is generally recommended to perform the final merge in image space.
|
||||
|
||||
- **Q: Why do I get multiple images as output even though I connected the nodes?**
|
||||
|
||||
- A: Ensure you are using `Advanced Tile Merger` and have correctly connected `tile_config`. If the output is still a list, check if you have connected an old node that does not support List after the Merger.
|
||||
|
||||
## 🤝 Contribution and Feedback
|
||||
|
||||
Feel free to submit Issues or Pull Requests. If you find areas for improvement while using this project, please reach out.
|
||||
|
||||
## 📜 License
|
||||
|
||||
This project is open-source under the [Apache-2.0 license](https://github.com/QL-boy/comfyui-ps-plugin#Apache-2.0-1-ov-file).
|
||||
|
||||
+128
@@ -0,0 +1,128 @@
|
||||
# ComfyUI-Advanced-Tile-Processing
|
||||
|
||||
**ComfyUI-Advanced-Tile-Processing** 是一套专为 ComfyUI 设计的高级自定义节点插件。它旨在解决超高分辨率图像生成和处理中的显存瓶颈(VRAM OOM),通过智能分块(Tiling)与基于加权累积的无缝合并(Merging)技术,让您能够在普通消费级显卡上稳定处理 4K、8K 甚至更高规格的图像。
|
||||
|
||||
## 🌟 核心特性
|
||||
|
||||
- **智能维度感知 (Smart Dimension Awareness)**:自动计算最优切片坐标。支持“边缘回退”策略,确保分块完全覆盖原始图像,无需引入黑边(Padding)从而避免模型产生伪影。
|
||||
- **无缝加权融合 (Soft Blending)**:内置多种混合模式(Gaussian, Cosine, Linear)。通过多层重叠区域的加权平均,最大化消除物理拼接产生的“网格接缝”。
|
||||
- **元数据驱动流 (Metadata-Driven Workflow)**:Splitter 生成唯一的 `TILE_CONFIG` 对象,包含原始尺寸、缩放因子及坐标。Merger 自动读取配置一键还原,用户无需手动对齐参数。
|
||||
- **深度兼容性**:完美适配 ComfyUI 的列表执行机制。支持将分块作为 `BATCH` 提交以获得最高推理速度,或作为 `LIST` 提交以支持循环节点处理。显存不足的情况下推荐 `LIST` 替代 `BATCH` 。
|
||||
- **高性能架构**:完全基于 PyTorch 张量运算,GPU 加速,保证处理处理大图时性能在线。
|
||||
- **动态尺寸支持 (Tiled Upscale)**:支持动态尺寸调整(Tiled Upscale),在分块后进行放大处理。Merger 能够自动检测片段尺寸变化并动态调整画布大小。
|
||||
|
||||
## 🛠 安装说明
|
||||
|
||||
1. **环境要求**:ComfyUI 0.4.0+,Python 3.10+,PyTorch 2.0+。
|
||||
|
||||
2. 进入您的 ComfyUI 自定义节点目录:
|
||||
|
||||
```
|
||||
cd ComfyUI/custom_nodes/
|
||||
```
|
||||
|
||||
3. 克隆本仓库:
|
||||
```
|
||||
git clone [https://github.com/QL-boy/ComfyUI-Advanced-Tile-Processing.git](https://github.com/QL-boy/ComfyUI-Advanced-Tile-Processing.git)
|
||||
```
|
||||
|
||||
4. 重启 ComfyUI。
|
||||
|
||||
|
||||
## 🧩 节点详解
|
||||
|
||||
### 1. 🔧 Advanced Tile Splitter (分块器)
|
||||

|
||||
将输入的大图或 Latent 空间分割成重叠的小块。
|
||||
|
||||
- **输入端口**:
|
||||
|
||||
- `image`: (可选) 原始图像。
|
||||
- `latent`: (可选) 潜在空间数据。
|
||||
|
||||
- **核心参数**:
|
||||
|
||||
- `tile_size`: 分块分辨率(如 512, 1024)。
|
||||
- `overlap`: 两个分块之间的重叠像素。建议设为 `tile_size` 的 10% 及以上以获得最佳融合效果。
|
||||
|
||||
- **输出端口**:
|
||||
|
||||
- `tiles_image_batch`: 将所有分块合并为一个 Batch Tensor,适合高性能采样。
|
||||
- `tiles_image_list`: (List 模式) 分块图像列表,触发 ComfyUI 循环执行。
|
||||
- `tiles_latent_batch / list`: 对应 Latent 空间的输出。
|
||||
- `tile_config`: 核心配置文件(必需连接至 Merger),包含合并所需的所有元数据。
|
||||
|
||||
### 2. 🔧 Advanced Tile Merger (合并器)
|
||||

|
||||
根据分块配置将处理后的片段无缝还原。
|
||||
|
||||
- **输入端口**:
|
||||
|
||||
- `tile_config`: 由 Splitter 输出的配置对象。
|
||||
- `processed_tiles_image`: (可选) 处理后的图像片段。
|
||||
- `processed_tiles_latent`: (可选) 处理后的 Latent 片段。
|
||||
|
||||
- **核心参数**:
|
||||
|
||||
- `blend_mode`:
|
||||
- `gaussian` (默认): 中心权重高,边缘平滑衰减,融合效果最自然。
|
||||
- `cosine`: 经典的三角函数衰减,过渡区域较宽。
|
||||
- `linear`: 线性金字塔混合。
|
||||
- `none`: 硬边缘拼接(会有明显接缝)。
|
||||
- `feather_percent`: 羽化比例 (0-50%),控制权重衰减的起始位置,值越大,边缘融合区域越宽。
|
||||
|
||||
- 内部机制:
|
||||
|
||||
节点会自动解包 `LIST` 输入。如果您的上游是循环节点,请确保所有分块均已处理完成再输入 Merger。
|
||||
|
||||
## 📖 核心算法原理:加权融合(Soft Blending / 累积缓冲加权平均法)
|
||||
|
||||
为消除传统“剪切-粘贴”方式产生的接缝与亮度突变,本插件采用**累积缓冲加权平均法(Accumulation Buffer Weighted Averaging)**,实现多层像素在重叠区域的平滑融合。其核心流程如下:
|
||||
|
||||
1. **掩码生成(Weight Mask Generation)**
|
||||
为每个图像分块(Tile)生成一个权重掩码矩阵 $M$。在 Gaussian 模式下,中心点权重为 1.0,并向边缘按指数规律衰减至接近 0,确保融合时中心区域贡献最大、边缘逐渐过渡。
|
||||
2. **像素累加与权重累加**
|
||||
- **像素累加**:在全局画布对应位置累加当前分块加权后的像素值:
|
||||
$Canvas += Tile \times M$
|
||||
- **权重累加**:同步在权重分布图(Weight Map)中累加掩码值:
|
||||
$WeightMap += M$
|
||||
3. **归一化输出(Normalization)**
|
||||
所有分块累加完成后,对每个像素进行归一化计算,得到最终图像:
|
||||
$Pixel_{final} = \frac{Canvas}{WeightMap + \epsilon} = \frac{\sum (Tile_i \times M_i)}{\sum M_i + \epsilon}$
|
||||
其中 $\epsilon$ 是为避免除零而设置的极小常数。
|
||||
### 关键优势
|
||||
- **无缝过渡**:重叠区域由多个 Tile 按权重共同贡献,通过归一化保持亮度与对比度一致;
|
||||
- **抗伪影**:权重平滑变化避免硬边缘,实现视觉上的自然渐变;
|
||||
- **灵活性**:可通过调整权重衰减曲线适应不同分块重叠策略。
|
||||
|
||||
该方法本质上是一种**软混合(Soft Blending)**,通过在重叠区域进行多图层像素的加权平均,达成高质量、无接缝的大图合成效果。
|
||||
## 🚀 示例工作流
|
||||
### SDXL 经典分块采样高清修复复刻
|
||||

|
||||
|
||||
### Z-Image质量升级
|
||||

|
||||
|
||||
###
|
||||
|
||||
## ⚠️ 常见问题 (FAQ)
|
||||
|
||||
- **Q: 为什么合并后的图边缘还是有淡淡的印子?**
|
||||
|
||||
- A: 尝试增加 `overlap` (建议不低于 64) 并确保 `blend_mode` 设为 `gaussian`。同时,检查重绘时的 `denoise` 是否过高,过高的降噪会导致分块内容产生剧烈变化。
|
||||
|
||||
- **Q: 支持 Latent 空间合并吗?**
|
||||
|
||||
- A: 支持。但由于 VAE 编码器的特性,Latent 空间的重叠合并可能会在 Decode 后出现微小色差,通常推荐在 Image 空间执行最终合并。
|
||||
|
||||
- **Q: 为什么我连接了节点却输出了多张图?**
|
||||
|
||||
- A: 请确保您使用的是 `Advanced Tile Merger` 并正确连接了 `tile_config`。如果输出依然是列表,请检查是否在 Merger 之后又连接了不支持 List 的旧版节点。
|
||||
|
||||
## 🤝 贡献与反馈
|
||||
|
||||
欢迎提交 Issue 或 Pull Request。如果您在使用中发现本项目可以改进的地方,请随时联系。
|
||||
|
||||
## 📜 许可说明
|
||||
|
||||
本项目基于 [Apache-2.0 license](https://github.com/QL-boy/comfyui-ps-plugin#Apache-2.0-1-ov-file) 许可证开源。
|
||||
Reference in New Issue
Block a user