Making Original Character Process

In this entry, I’ll share my workflow for creating an original character for image-generation projects.

Tools: GPT 5.5 or 5.6 sol & use ChatGpt project feature
Context: Lore.md, Character.md, Rule.md, Style.md
Skill: Image-to-image.md, QC.md

----
Context engineering structure

CHD-Project/

├── Context/ 

│ ├── Lore.md 

│ ├── Character.md 

│ ├── Rule.md 

│ └── Style.md 

├──  Skill/ 

│ ├── Image-to-image.md

│ └── QC.md

├── Referrence 

│ └── Ref.png (s)

----

Step 0 : Design the Character

> Define the character’s core concept:

  • Basic information
  • Physical appearance
  • Personality
  • Behavior
  • Motivations, strengths, and weaknesses
  • Other defining characteristics

> You should have a crystal-clear image of the character in your mind before proceeding.

Step 1 : Setting up.

Upload the Context Files

  • Lore.md — Worldbuilding, setting, mood, and themes
  • Character.md — Universal rules and standards for character design
  • Rule.md — Essential project rules and constraints
  • Style.md — Art style, lighting, color palette, and visual direction
  • Reference images (.png)

Upload the Skill Files

  • Image-generation.md — Adapted from an image-generation skill in the Skill Marketplace
  • Image-to-image.md — Instructions for making precise changes to existing images
  • QC.md — Quality-control and acceptance criteria


Step 2 : Create 'Character Bible' for the OC

  • Write a detailed character fact sheet.
  • Develop the character’s background story.
  • Identify the essential information required to preserve the character’s identity and consistency.
  • Separate fixed canonical details from optional or flexible details.


    • Step 3 : Command AI to review the Context and generate character sheet

  • Ask the AI to review all uploaded context and skill files.
  • Ask it to summarize its understanding of the project.
  • Provide the essential information about the character.
  • Ask the AI to identify any missing or unclear character details.
  • Generate the character fact sheet and visual character sheet.
  • Evaluate the result using the QC criteria.
  • Refine and edit the output until it meets the acceptance criteria.



    • Step 4 : Geneate Character 'Refined raw material'

    In my view, this is a crucial step because it has a significant impact on character consistency across the generated images.

    1. Upload the approved character sheet.
    2. Generate reference images of the character from multiple angles and perspectives.
    3. Include character-specific negative prompts to prevent recurring errors or unwanted variations.
    4. Evaluate the results using the QC criteria.
    5. Refine the selected images and apply any necessary minor edits.
    6. These refined images will serve as the primary reference material for subsequent image generation.




    Step 5 : Quality upgrade and Resolution enhancement

    1. Import the refined reference images into a digital art application, such as PaintTool SAI.
    2. Manually correct any remaining visual errors, such as anatomical distortions, malformed details, or extra limbs.
    3. Clean up the linework, colors, and other important character details.
    4. Enhance the image resolution while preserving sharpness and visual consistency.
    5. Perform a final quality check before approving the reference images.

    The character is now ready to be placed in specific scenes, actions, and situations.

    ----

    Example showcases






    Image-to-Image Logic-prompt; Trapped benated cage underwater.


    Image-to-Image Logic-prompt; Trapped benated cage underwater-Full body


    Image-to-Image Logic-prompt; Trapped situation underwater


    Image-to-Image Logic-prompt; into the sewer.


    Comments

    Post a Comment