Introduction

OpenPose, a widely-used protocol for defining human figure poses, provides a standardized method to represent humanoid joints and connections using images. This article explores how to leverage OpenPose within ComfyUI for creating text-to-image workflows. By combining OpenPose ControlNet models and pose images generated externally, users can define precise humanoid poses for AI image generation.

This guide will walk you through setting up OpenPose in ComfyUI, configuring the necessary components, and troubleshooting common issues, including problems with the OpenPose Editor node in the latest desktop version.

Setting Up OpenPose in ComfyUI

Step 1: Download the Required Components

To use OpenPose in ComfyUI, you need two critical components:

OpenPose ControlNet Model: This model is tailored for the specific diffusion model you’re using. For SDXL users, download the controlnet-openpose-sdxl-1.0 from Hugging Face. Locate the required file under the “Files and Versions” tab and place it in your ComfyUI installation directory under Models > ControlNet.
OpenPose Editor Node: Download and install the ComfyUI-OpenPose-Editor node. Use the ComfyUI manager to search for and install this component. Restart ComfyUI after installation.
Pose Image Generation Service (Optional): If you’re facing issues with the OpenPose Editor node in the latest desktop version of ComfyUI, you can use an online OpenPose image generation service Openpose Editor to create pose images. Then download the pose image and load post image into ComfyUI using the Load Image node.

Step 2: Build Your Workflow in ComfyUI

Workflow Setup

Load ControlNet Model: Add the Load ControlNet Model node to your workflow and load the OpenPoseXL2 ControlNet model. Connect it to the Apply ControlNet node, which provides conditioning for the Sampler node.
Load Pose Image: Instead of using the OpenPose Editor node, add a Load Image node to your workflow. Use this node to import the externally generated pose image. Connect the image output of the Load Image node to the image input of the Apply ControlNet node.
Prompt: Fashion photography, isolated figure of a 17-year-old girl dancing ballet, 50’s retro vibe, graceful expression, flowing hair, wearing a white ballet leotard, sheer wrap skirt, pink ballet slippers, vibrant pose, mid-pirouette, dynamic and elegant movement.

Step 3: Generating Pose Images Externally

Why Use External Services?

The OpenPose Editor node in the latest desktop version of ComfyUI has known issues, including limited functionality and display problems. To bypass these issues:

Use an online OpenPose image generation service Openpose Editor to create pose images interactively. These services allow you to define humanoid poses without relying on the faulty editor.

Steps to Generate Pose Images Online

Visit an online OpenPose image generation platform.
Create and customize your pose interactively.
Download the generated image in PNG format.
Import the image into ComfyUI using the Load Image node.

Step 4: Optimizing Workflow Parameters

Adjust ControlNet Application

To improve performance:

Lower the end_percent parameter in the Apply ControlNet node to 0.25. This ensures the ControlNet is applied only during the first 25% of the generation steps, speeding up the process while maintaining accuracy.

Queue the Workflow

After setting all parameters, queue the workflow and generate your image.

Troubleshooting Common Issues

Issue 1: OpenPose Editor Node Problems

Problem: The OpenPose Editor node in the latest desktop version of ComfyUI has display and functionality issues. It’s still not fixed after I tried all the possible solutions from internet searching.
Workaround: Use an online OpenPose image generation service Openpose Editor to create pose images and import them into ComfyUI using the Load Image node.

Issue 2: Canvas Visibility in OpenPose Editor

Problem: The editor window may not display properly in the browser-based version of ComfyUI.
Solution: Zoom out using Ctrl + Mouse Wheel in your browser to reveal hidden controls.

Issue 3: Facial Orientation and Expression

Problem: Poses with specific facial orientations may not render as desired.
Solution: Use inpainting techniques or re-roll with a different random seed to refine facial details.

Workflow:

Conclusion

By integrating OpenPose within ComfyUI, users can create detailed humanoid poses for AI image generation workflows. While the OpenPose Editor node in the latest desktop version has known issues, using external services to generate pose images offers a reliable alternative. Importing these images via the Load Image node ensures seamless workflow execution.

For advanced refinements, techniques like inpainting can further enhance outputs. For more information, refer to the official documentation on Hugging Face and the ComfyUI GitHub repository.

C24: How to Use OpenPose in ComfyUI for Humanoid Pose Creation in AI Workflows