1. Workflow Overview

This workflow generates Chinese-style anime novel covers using MR Anime Gufeng model with Flux sampling, achieving ink-wash effects and dynamic compositions. It features dual-CLIP encoding, flux guidance control, 4x upscaling, and automatic metadata embedding.

2. Core Models

3. Critical Nodes

• FluxGuidance

  • Param: guidance=3.5 (controls composition dynamism)

  • Effect: Generates high-angle perspectives and dramatic lighting

• ModelSamplingFlux

  • Key Param: max_shift=1.15 (maximum composition offset)

  • Dependency: comfyui-flux-diffusion plugin

• easy showAnything

  • Built-in Chinese-style prompt template

  • Customizable: Character traits (default: water sprite boy)

4. Pipeline Structure

graph LR
    A[Dual-CLIP] --> B[Flux Guidance]
    B --> C[Flux Sampling]
    C --> D[4x Upscale]
    D --> E[Metadata Export]

5. I/O Specification

Input Requirements:

• Fixed Resolution: 824×1280 (portrait ratio)

• Seed: Auto-randomized (e.g. 786693376160728)

• Prompt: Auto-loaded ink-wash template

Output:

• Upscaled PNG (actual size 1648×2560)

• Embedded Metadata: Prompt/Seed/Model info

6. Important Notes

⚠️ Mandatory Setup:

git clone https://github.com/Acly/comfyui-flux-diffusion  # Flux dependency

⚠️ VRAM Requirements:

• Base Generation: ≥6GB

• Upscaling: ≥10GB
  ⚠️ Key Features:

• Uses uni_pc_bh2 sampler (fast convergence)

• Auto-side-margins (for book binding)