A collection of ControlNet poses designed to enhance control in text-to-image diffusion models. It offers a curated dataset of diverse human poses, each accompanied by multiple conditioning map variants (OpenPose, Canny, Depth, Normal). This resource empowers users of tools like Stable Diffusion with ControlNet extensions to dictate the structure, silhouette, and perspective of generated images, enhancing creative iteration and prototyping.

How It Works

The project provides image collections where each pose is pre-processed into several conditioning formats. OpenPose skeletons isolate pose structure, Canny maps define outlines, Depth maps convey spatial relationships, and Normal maps detail surface orientation. By combining these, users can guide diffusion models to generate images adhering to specific poses and compositions, offering a powerful layer of control beyond simple text prompts.

Quick Start & Requirements

Pose collections are available for direct download from GitHub Releases. Usage typically involves integrating these maps with extensions like ControlNet within Stable Diffusion environments (e.g., Stable Diffusion web UI, Stability Matrix). Specific hardware requirements are dictated by the underlying diffusion models. A WIP webpage is available for browsing and filtering poses.

Highlighted Details

• Provides OpenPose, Canny, Depth, and Normal map variants for each pose.
• Direct download access via GitHub Releases.
• WIP web interface for interactive pose exploration and filtering.
• Includes example generation parameters for reference.

Maintenance & Community

Contributions via GitHub Issues (bugs, pose requests) and Pull Requests are encouraged. Direct contact is available via the author's GitHub profile email. Starring the repository ensures timely release notifications.

Licensing & Compatibility

Released under the Apache License 2.0, which permits broad commercial use and integration into proprietary software.

Limitations & Caveats

The dedicated web browsing interface is explicitly marked as a Work In Progress (WIP). The README does not detail performance benchmarks or specific model compatibility beyond general ControlNet usage.