ComfyUI Extension: comfyui-pose-redirect-align
Run ComfyUI workflows without the setup
No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.
Align one pose image to another by matching subject scale and head position for ComfyUI.
README
ComfyUI Pose Redirect Align
This custom node aligns source_pose to the composition of reference_pose.
What it does:
- Detects the person foreground area in
reference_pose. - Detects the person foreground area in
source_pose. - Scales
source_poseso its person height matches the person height inreference_pose. - Estimates a head anchor in both poses using the upper body region.
- Translates the scaled
source_poseso the head position matches the reference pose.
Outputs:
aligned_pose: the remapped pose image, ready for ControlNet.aligned_mask: the foreground mask after alignment.scale: applied scale factor.offset_x/offset_y: applied translation.
Install
Copy the whole comfyui_pose_redirect_align folder into:
ComfyUI/custom_nodes/
Then restart ComfyUI.
Node Name
姿态重定向对齐
新增姿态数据版节点:
姿态数据重定向对齐
这个节点直接处理 pose_metas_original 数据:
- 只用双方共同存在且置信度足够高的身体关键点求主变换
- 缺失点不参与对齐求解
- 源姿态中已有的点会保留,并统一映射到参考姿态画布
- 优先使用肩胯躯干,其次退化到肩线,再退化到头部
Category:
pose/redirect
Remote Deployment
Important: a custom node must be installed on the machine that is actually running ComfyUI.
- If you control the remote server, upload this folder to the server's
ComfyUI/custom_nodes/directory and restart ComfyUI. - If the remote platform does not allow custom nodes, this node cannot be used there.
- If the remote platform has a custom node manager, package this folder as a zip or git repo and install it from the server side.
Tuning
- Increase
background_thresholdif the pose image background is not clean. - Adjust
head_search_ratioif the head anchor is detected too high or too low. - Widen
min_scaleandmax_scaleif the two poses differ a lot in size.
Run ComfyUI workflows without the setup
No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.