Dane DeSutter, Ph.D.
Try it live

Mannequin Model

Pose a wooden mannequin by hand, or by describing the pose.

The Mannequin Model app: a wooden artist's mannequin shown mid-stride in a walking pose on a dark floor, beside a control panel with a text box reading Walking, arms swinging, suggested poses, and buttons for each joint group.
The mannequin in a walking pose, generated from the text “Walking, arms swinging.”

Mannequin Model is a 3D artist’s wooden mannequin that runs in your browser. You can pose it by hand, one joint at a time, or describe the pose in plain words, like “walking, arms swinging,” and have Claude set the joints for you. Each of its 18 joints is limited to a range a real body could reach, so the poses stay believable. When you’re done, you can copy the pose as YAML or save a picture of it.

Why I built it

Image generators often get bodies wrong. I built the mannequin so a pose can ground image generation: pose it first, then use the pose to guide a diffusion model, so the people in the generated images come out with accurate bodies. It connects to Body Talk, where the body is the subject.

Where it stands

The posing tool works, and the code is public on GitHub. The step that hands a pose to an image model is not part of the demo, so the demo does not generate images.

Try it

The posing tool runs in your browser, and you can dial in any of the 18 joints by hand. The text-to-pose generator needs an Anthropic API key, so it is turned off on the hosted demo.

Open the live demo ↗