Openpose – Joints Master ControlNet

Hello everyone! In my last post, I briefly explained ControlNet’s Openpose, and today, I want to delve a bit deeper into the fantastic dw_Openpose.
While the original Openpose extracts the skeleton from a photo to dress our desired model, it could only capture the larger bones and the face. Now, with dw_openpose_full, we can capture even the intricate details of finger bones! Exciting, isn’t it?

Openpose?

OpenPose is a real-time multi-person human pose detection library, which was the first to simultaneously detect human body, foot, hand, and facial keypoints in single images. I’m going to show you how to use Openpose more deeply than what we experienced in the previous posting. Shall we start?

1. Preparations

• ControlNet Extension
• Openpose Model
• Images showing fingers and face (any image is OK)

First, if you’re new to ControlNet, you can install it using the link below.

–> Settings for ControlNet

–> Coy and Paste it in Automatic1111-Extension Tab’s Url Install : https://github.com/Mikubill/sd-webui-controlnet.git

For those who need models or processors for Openpose, you can download them from the link below and place them in the sd-webui/Extension/ControlNet/Model folder.

–> Model Download Link

2. Openpose & Openpose_full

With all preparations complete, let’s first look at the differences between basic Openpose and dw_Openpose.
Scroll down to the ControlNet tab without doing anything else.

First, we’ll look at the processor models Openpose and Openpose_full.

Drag and drop the photo with a clear face and fingers into the image space, select Allow Preview, then choose Openpose from the dropdown, set Preprocessor to Openpose, and select control_v11p_sd15_openpose as the Model.

Press the red spark button in the middle. When the red button is pressed, a loading screen will appear next to the image, and the extracted skeleton will be visible on a black screen.

Openpose, as shown above, only extracts major bones, making it useful for separate editing.
However, it’s challenging to replicate the exact appearance of the photo.

Next, change the Preprocessor to Openpose_full and press the red spark button again.

With Openpose_full, as you can see, it builds a skeleton including the eyes, nose, mouth, and fingers from the original image.

However, if the fingers in the photo are somewhat complex, the bones may not be extracted correctly, which can be problematic for the work. Of course, you can edit these joints, but for those who want simple usage, I recommend using dw_openpose, which I will explain next.

3. DW_Openpose_Full

I’m sure all AI artists who struggled with complex finger positioning and rendering would have shed tears of joy with this preprocessor.
As explained earlier, change the Preprocessor to dw_openpose_full and press the red spark button to check.

Isn’t it incredible?? It seems to perfectly extract the skeleton from the photo I put on the left.
Now, based on this extracted skeleton, let’s write the model and prompts to generate an image!

4. Generate Image with dw_openpose_full

For those who find it challenging to start from scratch, please follow along with the settings I specified.

Settings

• Base Model : reallife_v10.safetensors
• Vae : Automatic
• Positive Prompt : masterpiece, best quality, 8k, uhd, (ultra realistic, realistic, highly detailed, photo realistic) 1girl, purple hair, off-shoulder fur jacket, purple and pink nail arts, sunlight, paled skin
• Negative Prompt : ng_deepnegative_v1_75t, (worst quality, low quality), watermark, monochrome, nude
• Sampling Method : 4x-UltraSharp
• Sampliong Steps : 30
• Hires.fix / Upscaler : 4x-UltraSharp
• Hires.fix / Upscale by : 2
• Hires.fix / Hires Steps : 20
• Desnoising Strength : 0.2
• Size : 512 x 960
• CGF Scale : 7
• Seed : -1

Now that all the settings are complete, go back to the ControlNet window and make sure to activate the Enable checkbox! Then, press the Generate button. Just seeing the finger and facial joints is exhilarating, and I’m extremely curious to see what kind of result we’ll get!

Can you see it? This is truly revolutionary. While it might be a combination of existing technologies, the fact that now anyone can easily create stunning artworks, just with the desired pose data, is absolutely astounding.

I encourage you all to not just stick to the settings I’ve used, but also experiment with different prompts or LoRA data to create your images!

Next time, I’ll bring even more new and exciting content! Stay tuned~!