liveBook · Manning

Pose

As explained in Local Image Generation

Pose structure is the spatial arrangement and configuration of a subject's body, face, and hands represented for image conditioning.

Pose is the spatial arrangement of a subject’s body, face, and hands. Prompting can usually manage broad actions such as sitting or standing, but it struggles when you need an exact configuration. Open Pose supplies that missing blueprint for image generation. section 12.4

A practical workflow starts with a reference image in ComfyUI. The DWPose Estimator extracts a pose structure, producing a general body layout plus detailed face and hand information. Feed that result to an Open Pose ControlNet, then generate a new subject while preserving the selected arrangement. section 12.4.1

Use the extracted pose selectively:

  • Keep body pose, face, and hands when the new character should closely match the reference. section 12.4.1
  • Disable facial or hand details when you want a different expression or do not need those features reused. section 12.4.1
  • Treat pose variety as useful training data: images showing different angles, distances, activities, and positions help a subject appear in varied poses. section 13.2.1

From statue to cyber ninja

Take a photograph of a statue of Patanjali meditating, extract its pose with DWPose, and use Open Pose ControlNet to generate a futuristic cyber ninja meditating in a garden with the same configuration. section 12.4 section 12.4.1

Figure 12.19 Statue of Patanjali meditating (attribution: Wikipedia user Alokprasad)

Inspect the meditation statue used as the reference before extracting its pose. section 12.4

How Local Image Generation covers it

How deep it goes
Explains reference extraction, component toggles, and ControlNet reuse, but stops short of low-level model internals.
Examples
Uses a meditating Patanjali statue, a cyber ninja generation, and varied subject-training images.
What it assumes
Assumes a source image and an image-generation workflow, while introducing pose extraction and component selection through a concrete task.
What you can do after
Apply it via ComfyUI, DWPose Estimator, and Open Pose ControlNet to transfer selected pose details into generated images.
How it connects
Links pose to prompting, ControlNet, ComfyUI, DWPose, Stable Diffusion, and LoRA training.

Also explained in

Compare how 3 books explain pose →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Local Image Generation. The section links point to the parts of the book it draws on.