AnyPose

Introduction

AnyPose helps you easily transfer poses to characters in Qwen Image Edit (2511) or its variants. Forget struggling with OpenPose's depth issues or distortions; AnyPose lets you skip it entirely. Just use a simple prompt and a reference image to copy any pose you want, no ControlNet needed. It's built to work fast, especially with the Qwen Image Edit 2511 lightning LoRA.

How to Use

• Trigger word: None required.
• Recommended settings:
  • Use the 4-step lightning LoRA for Qwen Image Edit 2511.
  • Set both AnyPose LoRA weights (base & helper) strength to 0.7.
• Steps:
  1. Upload two images: one with your initial character (Image 1) and another with the pose you want to copy (Image 2).
  2. Use this prompt: "Make the person in image 1 do the exact same pose of the person in image 2. Changing the style and background of the image of the person in image 1 is undesirable, so don't do it. The new pose should be pixel accurate to the pose we are trying to copy. The position of the arms and head and legs should be the same as the pose we are trying to copy. Change the field of view and angle to match exactly image 2. Head tilt and eye gaze pose should match the person in image 2."
  3. Tips and Tricks:
     • If you need to add context (like clothing not visible), append it to the end of the prompt.
     • If the background changes and you want to keep Image 1's background, append this: "Remove the background of image 2, and replace it with the background of image 1."
     • For best results, make sure the initial character's full body and environment are visible in Image 1.
     • You can append prompts to fix other issues, like adding flooring if the character is floating, or removing unwanted objects.

FAQ

Q: What's the trigger word?

A: No trigger word required.

Q: What LoRA strength should I use?

A: Set both AnyPose LoRA weights (base & helper) strength to 0.7. Feel free to experiment with other values!

Q: Can AnyPose handle complex poses?

A: It works well with many complex poses, like yoga. However, extremely technical or hyper-complex poses might still be a challenge.

Q: What if the background changes after applying the pose?

A: If you want to keep the original background from Image 1, just append this to your prompt: "Remove the background of image 2, and replace it with the background of image 1."

Q: What if my initial character image only shows part of their body?

A: AnyPose will try to fill in the missing areas by blending with the original image. For better consistency, it's best to use an initial image where the character's full body and environment are visible. You can also add context to the prompt (e.g., "The woman in image 1 is wearing white leggings.") to guide it.

Q: Does AnyPose work with 2D art styles like pixel art or cel-shaded?

A: Not really. AnyPose was trained mainly on 3D poses, so it struggles with 2D or flat art styles.

Q: Can I use AnyPose with multiple characters in an image?

A: No, AnyPose gets confused if there's more than one character in either the input image or the reference pose.

Technical Details

• Base model: Qwen/Qwen-Image-Edit-2511
• Training info: Trained using 3D poses of models created in Blender.
• Source: Hugging Face