A Guide to Creating Custom Character Art Using Multiple LLMs

Posted by:

|

On:

|

, ,

Step 1: Setting Up Your Environment

  • Install Open WebUI with GPU support on your desktop computer
  • Load the Qwen 2.5 14B model
  • Load the DeepSeek-R1 14B model
  • Ensure access to the Midjourney AI web explorer for reference

Step 2: Research and Reference Phase

  • Study successful Midjourney examples
  • Document effective keywords
  • Use the “Describe” feature for reference images
  • Build style descriptor collection (e.g., “glossy,” “smooth lines,” “circular patterns”)

Step 3: Building Your Base Prompt

  • Structure: [Image Type] + [Subject] + [Characteristics] + [Attire] + [Environment] + [Parameters]
  • Component Details:
    • Image Type/Style (select one):
      • “photo of…”
      • “product shoot of…”
      • “ink-wash illustration of…”
      • “3D…”
      • “hyper-detailed portrait of…”
    • Primary Subject:
      • Main subject description
      • Key defining characteristic
    • Distinctive Characteristics (max 6):
      • Physical attributes
      • Specific details
      • Notable features
    • Clothing/Attire:
      • Primary clothing description
      • Significant accessories
    • Environment:
      • Background setting
      • Lighting conditions
      • Atmospheric elements
    • Optional Mood:
      • Emotional tone
      • Visual atmosphere
  • Example Prompt:
    • 3D Anime of a chibi tiger-human hybrid, pristine white tiger body, elegant turquoise flowing hair, luminous pink eyes, crimson curved horns, majestic tail, traditional Chinese hanfu with flowing sleeves, dark gray desert background –ar 3:2 –profile x16lhyg –stylize 46

Step 4: Iterative Refinement Process

  1. Write your refinement query:

    “Using all the above information about Midjourney AI, modify the following prompt to [describe specific changes needed]:
    [paste original prompt]
    Important Prompt Principles:
    – Build clear visual hierarchy (main subject first, then details)
    – Balance details (guide without confusion)
    – Use natural language (like describing a real scene)
    – Focus on physical elements over abstract concepts
    – Create clear relationships between scene elements”

  2. Review responses from:
    • deepseek-r1:14b (left panel)
    • qwen2.5:14b (right panel)
  3. Now we have 2 Midjourney AI prompts at the same time

Step 5: Iteration and Optimization

Review Process: Generate the initial output, assess it against the objectives, and identify areas for improvement.

Refinement Cycle: If the results don’t meet objectives, pinpoint issues, adjust, and regenerate the output; otherwise, mark the process as complete and move to the next project.

This iterative approach ensures continuous improvement and successful outcomes.

Leave a Reply

Your email address will not be published. Required fields are marked *