Krea2 Turbo OpenPose ControlNet LoRA

Krea2 Turbo OpenPose ControlNet LoRA

A LoRA that makes Krea2 Turbo follow the body pose from an OpenPose or DWPose skeleton map while your prompt sets the look. For artists working in ComfyUI.

Made by thedeoxen 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 pose ControlNet-style LoRA for the Krea2 Turbo image model. You give it a skeleton map made with OpenPose or DWPose, and the generated figure follows that body pose. Your text prompt decides the look, clothing, style and scene.

Who it suits: Artists and designers who want control over how a character stands while still changing everything else by prompt. It was trained mainly on humans and also works on stylized characters.

Industry | Who uses it | How they use it | Example
Illustration | Illustrators | Fix a character's pose first, then vary outfit and style by prompt | Use one pose map with several different outfit prompts
Comics and storyboards | Storyboard artists | Keep body poses consistent while the scene changes | Reuse a pose map and describe a new location
Animation | Animators | Build keyframes from pose maps | Generate a keyframe image from a skeleton for each key moment
Concept art | Concept artists | Explore characters in a chosen stance | Pair a crouching skeleton with a prompt about a masked fighter

You can skip it if you want to blend a reference image with a pose. The card says this is not a reference plus pose fusion model.

Where to use it:

• ComfyUI, with the workflow file that ships in the repository
• Windows, macOS and Linux with Python, for downloading the files
• Image pipelines built on the diffusers library tag the page lists

Use case | Who | What they get back
Posed characters | Illustrators and comic artists | Images whose figures follow the pose map
Pose with free styling | Designers | The same pose rendered in different outfits and styles
Animation keyframes | Animators | Pose-driven frames for key moments
Stylized characters | Concept artists | Characters with controllable anatomy

How to set it up: This LoRA runs inside ComfyUI, so you need that and a few extras.

• ComfyUI installed and working
• The Krea2 Turbo base model, in the fp8 scaled file or an equivalent
• The comfyui_controlnet_aux nodes, which provide DWPose pose extraction
• The Krea2 Ostris Edit nodes, which allow conditioning with a control image

1. Download the repository files, which include the weights and the ComfyUI workflow.
2. Install the two node packs in ComfyUI.
3. Put the Krea2 Turbo base model where ComfyUI expects it.
4. Open the workflow file and load the LoRA weights.

pip install -U huggingface_hub
hf download thedeoxen/Krea-2-pose-controlnet

How to use it: Work through these steps in ComfyUI. Use a clean skeleton map, because the card says a black background works best.

1. Extract or load an OpenPose or DWPose skeleton map.
2. Give the map to the Krea2 Ostris Edit text encoder as its first image.
3. Write an ordinary prompt for the character and scene you want.
4. Attach the LoRA to the Krea2 Turbo UNet with a weight of 0.8 to 1.0.

Tool | What it does
DWPose | Extracts the skeleton map from a picture, through comfyui_controlnet_aux
Krea2 Ostris Edit text encoder | Takes the pose map as image 1 alongside your prompt
LoRA weights file | Teaches the model to follow the skeleton, loaded at 0.8 to 1.0
ComfyUI workflow file | A ready graph with about 10 steps, CFG near 1.0 and euler with simple

Example: You pair a skeleton of a crouching figure with a prompt about a black-suited night fighter holding two glowing blades on a rooftop in a rainy city. The image comes back with the figure in that crouch, dressed and lit as the prompt says. No trigger phrase is needed.

For newcomers and experts: A pose map is a stick figure drawn over a black background. This LoRA makes the image model copy that stick figure's stance. To start, run the included workflow with one pose map and a short prompt, and keep the LoRA weight near 1.0.

Under the hood it is a LoRA, not a separate ControlNet network. It teaches Krea2 Turbo to read the input image as a pose control map, so the subject lines up with the skeleton. Lower the weight to about 0.6 to 0.8 if the pose feels too rigid, and raise it toward 1.0 for tighter adherence.

Extract a DWPose skeleton map
Pass the map as image 1
Write a prompt for look and scene
Load the LoRA on Krea2 Turbo
Get an image that follows the pose

Good to know:

• It needs the ComfyUI node packs named above, plus the Krea2 Turbo base model.
• Keep the subject's scale and framing in the pose map sensible for the scene you want.
• Detailed prompts about lighting, camera angle and style combine well with it.
• License: apache-2.0.
• Cost to run: none named, though you need hardware that can run Krea2 Turbo in ComfyUI.
• Data it reaches: the pose images and prompts you supply, plus the files you download from Hugging Face.
• Not for: blending a reference image with a pose.

On Hugging Face, 13 November 2026

• Task: image to image
• Library: diffusers
• Downloads in the last 30 days: 18,552
• Likes: 133
• License: apache-2.0
• Based on: krea/Krea-2-Turbo
• Tags: diffusers, krea, krea2, openpose, controlnet, lora, comfyui, image-to-image

Links

• Model: https://huggingface.co/thedeoxen/Krea-2-pose-controlnet
• Files: https://huggingface.co/thedeoxen/Krea-2-pose-controlnet/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download thedeoxen/Krea-2-pose-controlnet

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