Design & Media: Creating Highly Realistic Characters and Consistent Narrative Scenes (Krea2 Workflow)

Creative Stack & Specs

• Core Toolset: Krea2 + ComfyUI orgetorally managed via Qwen VLM loop
• Target Medium: High-resolution realistic photography / cinematic character sheets
• Key Parameters:
  int8 model
  Wan2.1 fp32 VAE
  DDIM und Beta57
  Strength(Enhancer): 0.5
  Steps recommendation: -9, 11, 14
  Resolution targets: 704x1152 or 544x896/upscaled to 4k

Step-by-Step Production Workflow

1. Character Concepting: Develop an extremely detailed physical description for a subject including age, height, build, skin undertones, facial features, hair texture/style, eye shape, certain clothing textures, accessories, and even hand placement (e.g., A female investigative journalist... fair skin with warm undertone... shoulder-length naturally wavy dark auburn hair).
2. Prompt Separation & Scene Management: Use a main prompt structure that separates scenes using tags like SCENE 1 : [description] SCENE 2 : [description] enoughto ensure character consistency through extreme detail rather than low-res tiling.
   Use Qwen VLM in a loop inside ComfyUI to separate these into individual prompts per image if needed.
3. Image Generation / Base Pass: Execute generation using the Krea2 int8 model combined with DDIM und Beta57 settings at lower resolutions such as 704x1152 or 544x896 while implementing camera heuristics (shot on old iphone camera or old nikon 35mm camera 21.3MEGA PIXEL) or lighting instructions (HARSH SUNLIGHT, CONTRAST or soft even studio lighting).
4. Detail Enhancement (Upscaling): To correct lack of texture/skin pores often found in base models, apply two sample passes and use an upscaler like SeedVR2 for realistic skin textures으로sably producing high quality results or utilize any combination including Prompt Enhancer + 4K Upscaler via CivitAI metadata workflows.

Style Consistency & Quality Controls

• Character Continuity Strategy: Maintain near-perfect similarity across unlimited scenes by relying on hyper-detailed descriptive text blocks rather than traditional low-resolution panel tiling methods.
  (e.g., repeating exact physical specs [age / height / hair type] within every scene prompt block regardless of location change).
• Texture Management: Use a secondary pass with specialized enhancers to combat the

! DYOR (Do Your Own Research)