1. Workflow Overview

This "LibLib F1_CN_Union_Pro" workflow is a multi-ControlNet advanced pipeline for Stable Diffusion 1.5/XL, featuring four preprocessors (Canny/Depth/OpenPose/SoftEdge) to achieve precise image generation. Ideal for stylized illustrations, portrait reconstruction, and scene design.

2. Core Models

3. Key Nodes

4. Workflow Structure

1. Input Group

   • Input: Reference image (girl_portrait.png), prompts (auto-translated via Baidu node)

   • Process: ImageResizeKJ resizes to 1200x1600

2. Control Groups

   • Canny Group: Edge detection (weight=0.8)

   • SoftEdge Group: Line art control (weight=0.7)

   • Depth Group: Spatial structure (weight=0.7)

   • OpenPose Group: Pose control (weight=1.0)

3. Generation Group

   • KSampler: Euler sampler, 30 steps, CFG=3.5

   • LoRA Mix: Sketch style (0.8) + Flower style (0.35)

5. Input/Output

• Input Requirements:

  • Required: Reference image (PNG/JPG)

  • Optional: Prompts, seed (random by default), ControlNet weights

• Output:

  • 4 variant images with different controls

  • Saved to ComfyUI/output/%date%/

6. Critical Notes

1. Model Dependencies:

   • Must download FLUX.1-dev-ControlNet-Union-Pro-InstantX.safetensors to models/controlnet

   • Missing LoRAs will disable style effects

2. Performance:

   • VRAM ≥12GB recommended

   • Ref image resolution: 1024-1536px

3. Troubleshooting:

   • Red-box errors: Verify ControlNet model paths

   • Artifacts: Reduce ControlNet weights (0.6-0.8)