Shogo PSR Method

Pose Scene Reconstruction — An Experimental GPT-only Workflow

Shogo PSR Method is a lightweight experimental workflow for reconstructing the pose, camera, and composition of one image with a different character reference inside ChatGPT.

It started from a simple problem:

How can I use the pose and scene of one image while keeping the character, clothing, and visual style of another?

Instead of treating both images as general references, this method separates their roles before reconstruction.

The Idea

Pose Scene
    ↓
Observation
    ↓
Camera / Composition Normalization
    ↓
Character Authority
    ↓
Reconstruction

The Pose Scene provides the pose, action, camera, composition, scene geometry, and object interaction.

The Character Sheet provides the character identity, face, hair, clothing, accessories, colors, and ART_STYLE.

The goal is to prevent those two sources from unnecessarily exchanging roles.

Example

See examples/example-01-pitching/ for the initial experimental example:

Pose Scene Reference + Character Sheet → Shogo PSR Method → Generated Result

The initial pitching Pose Scene uses a CC0-licensed reference image.

This is an experimental example, not a benchmark or proof of general reproducibility.

How It Works

Step 1 — Camera Observation

A Pose Scene image and a short pose description are provided. Camera and composition characteristics are observed and recorded as CAMERA_OBSERVATION_DATA.

Step 2 — Normalization

The observation is converted into NORMALIZED_CAMERA_INPUT and SUPPLEMENTAL_COMPOSITION_CONSTRAINT.

Step 3 — Character Pose Generation

A Character Sheet is added. It becomes the authority for character identity, clothing, and ART_STYLE, while the Pose Scene remains the reference for pose, action, camera, composition, scene geometry, and object interaction.

Current Prompt

prompt/MFS_Pose_Scene_Reconstruction_Workflow_Prompt_v0.3.txt

It is published in its tested form rather than rewritten specifically for GitHub.

• Public Method: Shogo PSR Method
• Original / Internal Workflow: MFS Pose Scene Reconstruction (MFS PSR)
• Current Workflow Prompt: v0.3

What This Is — and Is Not

This repository shares one experimental way of structuring a pose-scene reconstruction problem.

It does not claim invention of pose transfer, character consistency, or GPT-only pose transfer; that this is the first or only workflow of its kind; higher accuracy than dedicated pose-control systems; exact pose reproduction; demonstrated general reproducibility; or patent novelty.

The interesting part of the experiment is the workflow:

Source Role Separation → Observation → Normalization → Character Authority → Reconstruction

Limitations

• pose fidelity is not exact
• left/right interpretation can fail
• joint positions or action phase can vary
• outputs can vary between runs
• only a small number of cases have been tested
• results may change with ChatGPT or image-generation model updates

Treat this as an experimental workflow, not a production pose-control system.

Repository Structure

Pose-Scene-Reconstruction/
├── README.md
├── LICENSE
├── THIRD_PARTY_NOTICES.md
├── method/
│   └── MFS_PSR_Method_v0.1.md
├── prompt/
│   └── MFS_Pose_Scene_Reconstruction_Workflow_Prompt_v0.3.txt
└── examples/
    └── example-01-pitching/
        ├── README.md
        ├── pose-reference.jpg
        ├── character-sheet.png
        └── generated-result.png

Background

Shogo PSR Method originated from the MFS (Morphological Face Scene) project. Understanding the broader MFS project is not required to use this workflow.

Status

Experimental / Lightweight Publication

Image Credits and Licensing

Third-party materials retain their original licenses and are identified in THIRD_PARTY_NOTICES.md. Repository-created materials are intended for release under CC BY 4.0 as described in LICENSE.