controlnet-openpose-sdxl-1.0

controlnet-openpose-sdxl-1.0

Controls the pose of people in Stable Diffusion XL images by feeding the model a body-pose skeleton drawing. For artists and developers who want figures to stand or move the way they choose.

Made by xinsir on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: This is a ControlNet add-on for Stable Diffusion XL, fine-tuned from the SDXL base model. It reads a body-pose skeleton image and steers the generated picture so the person matches that pose. The authors call it state of the art and show example images in styles like Midjourney and anime.

Who it suits: It suits artists and developers who generate images with SDXL and want to fix the pose of a figure instead of leaving it to chance.

Industry | Who uses it | How they use it | Example
Illustration | Digital artists | Draw a pose skeleton and generate a character in that pose | A character standing with arms raised
Games and animation | Concept artists | Keep a character's pose steady across several images | Variations of one fighting stance
Fashion and e-commerce | Visual teams | Pose a model figure the same way in different outfits | Showing a garment on a set pose
Software | Developers | Add pose control to an image generation tool | A pose input in a diffusers-based app

You can skip it if you do not generate images with SDXL, or if you do not need control over body pose.

Where to use it:

• Python code that uses the diffusers library with an SDXL ControlNet pipeline
• Windows, macOS or Linux, with Python for the download
• Pose preprocessing through the controlnet_aux package, which the card refers to

Use case | Who | What they get back
Generate a figure in a set pose | Artists | An SDXL image whose person matches the skeleton
Reuse one pose in many scenes | Designers | Several images with the same body position
Build pose control into an app | Developers | A pipeline that takes a pose image and a prompt

How to set it up:

• Python on your computer
• The diffusers library and an SDXL base model
• The controlnet_aux package for making pose images
• Disk space, about 2.5 GB for the half-precision weights

1. Install the Hugging Face hub package with pip.
2. Download the ControlNet files.
3. Load them in a diffusers SDXL ControlNet pipeline together with the SDXL base model.

pip install -U huggingface_hub
hf download xinsir/controlnet-openpose-sdxl-1.0

How to use it: You make a pose image from a reference picture, then give that image and a text prompt to the pipeline. The part of the card shown here begins its starting code with this import.

from diffusers import ControlNetModel, StableDiffusionXLControlNetPipeline, AutoencoderKL

The card also gives a tip. The default pose drawing in controlnet_aux uses thinner lines than the training data did, which can make results unstable. It supplies a replacement for the body-drawing function in that package's open pose utility file.

Example: A person draws or extracts a pose skeleton of someone with raised arms and writes a prompt for a character in a garden. The pipeline returns an image of that character with the same arm position. The card's replacement drawing function is meant to make this more stable.

For newcomers and experts: If you are new to this, start with one clear pose picture and a short prompt, and check that the output figure copies the skeleton. Keep the reference photo simple, with one person in view.

If you know the field, the card says the model was trained with thicker pose lines than the default drawing produces. It suggests patching the draw function in controlnet_aux so the conditioning image matches training. The replacement scales the line width with the image size.

Download the ControlNet files
Make a pose image from a reference
Load SDXL with the ControlNet
Add a text prompt
Generate the posed image

Good to know:

• Limits: the card gives no benchmark table, and its starting code is cut in the part shown here.
• Claims: the state of the art statement comes from the author.
• Pose drawing: the default controlnet_aux drawing can give unstable results until you swap in the card's function.
• License: Apache 2.0.
• Cost to run: none named.
• Data it reaches: it copies the files to your computer and reads the images you give it.

On Hugging Face, 14 November 2026

• Task: text to image
• Library: diffusers
• Parameters: 1.3B
• Weights: about 2.5 GB in F16; about 626 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 294,030
• Likes: 334
• License: apache-2.0
• Tags: diffusers, safetensors, openpose, controlnet, controlnet-openpose-sdxl-1.0, text_to_image, text-to-image

Links

• Model: https://huggingface.co/xinsir/controlnet-openpose-sdxl-1.0
• Files: https://huggingface.co/xinsir/controlnet-openpose-sdxl-1.0/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download xinsir/controlnet-openpose-sdxl-1.0

Then follow the usage in its model card. Nothing in this kit runs by itself.