DD Advanced Fusion
Blend two images with a line or a polygon, and live to tell about it
- 输入A
- 输入B
- 融合结果
DD Advanced Fusion is a compositing node with a specific flavor: instead of just fading A into B, it lets you carve the boundary with a straight line at any angle, or with a regular polygon, and it works on image batches too - so you can blend two videos frame-by-frame the same way. It's part of the DD-Nodes pack's image processing set, and it's the node in the collection with the most "wow, that's clever" energy and the most "wait, what does this do again?" confusion.
How it works
The core is a weighted mask blend: result = A * mask + B * (1 - mask). The 融合类型 (Fusion Type) chooses what that mask looks like:
- 单线融合 (Single-line blend) builds a diagonal seam through the image center.
融合角度(Fusion Angle) rotates that seam, and融合比例(Fusion Ratio) shifts where it sits - 0.5 splits the image exactly in half, lower values give B more of the frame. It's the "one image transitions to the other along a line" effect. - 多边形融合 (Polygon blend) draws a filled regular polygon in the middle.
多边形边数(Polygon Sides, 3–12) picks triangle/hexagon/whatever, the angle rotates it, and the ratio scales its radius. The polygon region shows image A; everything outside shows image B.
边缘模糊 (Edge Blur, 0–50) blurs the boundary with a Gaussian, which is what makes the seam look like a real blend instead of a hard mask. Turn it up when the cut looks surgical.
Two more settings handle mismatched inputs. 尺寸适配 (Size Fit) is the same adaptive/stretch/crop/fill family as the pack's resize nodes - B gets matched to A's dimensions. 帧数适配 (Frame Fit) decides what happens when A and B are videos of different lengths: shorter, longer, or average, with B's frames interpolated to match A's count. Handy if you're cross-fading clips of different durations.
When you'd reach for it
The classic use is a slideshow-to-video crossfade, or compositing a subject crop into a background where you want a soft geometric seam rather than a hard paste. Polygon mode is genuinely fun for stylized collages - a hexagonal window of one image inside another reads as intentional design, not a masking accident. If you just want a plain two-image fade, this still works, but honestly a simpler blend node is easier; the angle and polygon controls are the reason to pick this one.
Installation
Standard DD pack install:
cd ComfyUI/custom_nodes
git clone https://github.com/Dontdrunk/ComfyUI-DD-Nodes
cd ComfyUI-DD-Nodes && pip install -r requirements.txt
Restart ComfyUI, or use ComfyUI Manager and search "ComfyUI-DD-Nodes". No model downloads. The node leans on OpenCV under the hood for the mask work, which the pack already pulls in.
Where people get burned
The angle input has an odd edge - the schema allows 融合角度 down to 1 but defaults to 0, so a 0-angle single-line seam renders as a vertical cut in practice. Don't overthink the units; just drag the slider and watch the preview. Also, this node works in numpy/OpenCV, so big batches and high resolutions can get slow - keep 边缘模糊 modest on 4K content. And remember the fields are Chinese by default; the pack ships a full English locale if you switch ComfyUI's language. It's a niche tool, but when you need a rotated seam or a polygon blend, nothing else in the standard toolset does it as directly.
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| 输入A | IMAGE | — | |
| 输入B | IMAGE | — | |
| 融合类型 | COMBO | 2 options: 单线融合, 多边形融合 | |
| 融合角度 | INT | 01–360 | — |
| 融合比例 | FLOAT | 0.500–1 | — |
| 边缘模糊 | INT | 00–50 | — |
| 多边形边数 | INT | 63–12 | — |
| 尺寸适配 | COMBO | 4 options: 自适应, 拉伸, 裁剪, 填充 | |
| 帧数适配 | COMBO | 3 options: 较短, 较长, 平均 |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| 融合结果 | IMAGE | — |