1. Workflow Overview

This "Wan2.1 Preprocessor" workflow utilizes Flux Flill model to intelligently outpaint input images into square compositions. It combines Differential Diffusion with Flux Guidance for seamless edge transitions.

2. Core Models

3. Critical Nodes

• DifferentialDiffusion (Node 39):
  Dynamically adjusts edge generation intensity
  ▶ Requires Flux-series models

• Image Comparer (rgthree):
  Interactive before/after comparison UI
  ▶ Install via ComfyUI-rgthree extension

4. Workflow Logic

Data Preprocessing Group:

• Automatically processes all images in input directory

• Calculates padding dimensions through math nodes

Outpainting Group:

1. Generates padding mask via AI analysis

2. Performs latent space diffusion (20 steps)

3. Outputs 4K-ready square images

5. Key Parameters

• Input:

  • Source image directory (modifiable)

  • Seed: 764442076935121

• Output:

  • Saved to local path with metadata

6. Pro Tips

⚠️ Requirements:

• Minimum 8GB VRAM for 1024x1024 generation

• Correct model folder structure
  💡 Recommended:

• Use --medvram flag for GPUs <12GB

• Adjust KSampler scheduler per hardware