🐳YOLOv11图像粘贴

YOLOv11 Image Paste (🐳YOLOv11图像粘贴) is the V11 companion to the pack's paste-back node - and it fixes the one thing that bugged the v8 version: it adds color matching and edge blending, so refined crops actually sit in the original instead of floating on top of it. If you've ever pasted a re-rendered face back and watched the skin tone disagree with the neck, this is the node aimed at that exact seam.

Same skeleton as the v8 paste node: original_image, paste_images (a list input, so the V11 crop node's 裁剪图像 output wires straight in), and bboxes (the BBOXES output that records where each crop was cut from). Then paste_mode now offers four options - 全部粘贴, 指定索引, 循环使用, plus a new 智能混合 (smart blend) mode - with target_index for the index modes. feather_amount (0-100 px) and blend_alpha (0-1) work as before.

The new part is the pair of toggles that make compositing smarter:

• color_match (default on) - tries to reconcile the pasted patch's color/lighting with the surrounding original, so a re-rendered face doesn't read as a different color grade than the rest of the photo
• edge_blend (default on) - extra blending along the patch boundary, on top of the feather, to kill the hard edge

Both default on, which is the right call - those are the two things that make a paste look "AI-stuck-on" vs "composited". If you're pasting something that must stay exact (a logo, a texture) you can switch them off and get the plain hard paste.

Outputs add a third compared to v8: 拼接图像 (the composited IMAGE), 合成遮罩 (the MASK of pasted regions), and 处理信息 (a STRING describing what the paste did - useful for logging or debugging which crops went where).

How to use it well

The full loop is: YOLOv11 Smart Crop → process the crops (inpaint, detail, upscale) → this node pastes them back. The 智能混合 mode plus color_match on is the setting for re-rendered faces or body parts where tonal consistency is everything; 全部粘贴 with color matching off is the setting for pasting crops you deliberately didn't change. If your refined crops came out with very different brightness than the original, color matching does a real job of pulling them back toward the source - but don't expect it to fix a fully re-lit image. It's a statistical nudge, not relighting.

Honest limitations: same positional model as v8 - crops get resized to their bbox, so don't change aspect ratios upstream. And color_match is best-effort; extreme exposure mismatches will still show. This is a single-node convenience over the classic "paste + color match + blend" stack, and for the common face-detailing case it's genuinely the nicer button to press.

Install

Part of Pond Nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/Pondowner857/comfy_Pond_Nodes
cd comfy_Pond_Nodes
pip install -r requirements.txt

Restart after (or Manager → "comfy_Pond_Nodes"). Needs the pack's own BBOXES type from the V11 crop node - install both. Ultralytics is only needed on the crop side; this node is torch/PIL/OpenCV under the hood. Pack caveat: console spam if comfyui_HiDream-Sampler is installed alongside.