1. Workflow Overview

This ComfyUI workflow is designed for generating high-quality 6K wallpapers, combining text-to-image and image-to-image pipelines with:

• IPAdapter: Style consistency using reference images

• ControlNet (Canny edge): Preserves composition structure

• UltimateSDUpscale: Two-stage super-resolution upscaling

• Auto-negative prompts: Quality enhancement

2. Core Models

3. Key Nodes

4. Workflow Groups

Group 1: IPAdapter Style Reference

• Input: Style image (e.g., ComfyUI_temp_falxa_00007_.png)

• Output: Style-adapted model

• Key Nodes: IPAdapterModelLoader → IPAdapter

Group 2: Base Text-to-Image

• Input: Positive prompt (e.g., aurora whales) + auto-negative prompts

• Output: 1024x1024 image

• Key Nodes: KSampler + ControlNetApplySD3

Group 3: First Upscale

• Input: Base image + ESRGAN

• Output: 2K image

• Config: UltimateSDUpscale (tiles=4)

Group 4: Secondary Upscale (Optional)

• Input: 2K image + prompt refinement

• Output: 6K final wallpaper

• Note: Minimal quality gain per workflow Note node

5. Inputs & Outputs

• Primary Inputs:

  • Positive prompts (via PrimitiveNode or API)

  • Style reference image (upload to LoadImage)

  • Resolution preset (CR Aspect Ratio)

• Outputs:

  • 6K PNG (saved by SaveImage)

  • Comparison view (Image Comparer)

6. Notes

1. Hardware: Recommended 12GB+ VRAM (16GB for secondary upscale)

2. Troubleshooting:

   • IPAdapter model missing → Download to models/ipadapter

   • ControlNet errors → Verify control_v11p_sd15_canny.pth path

3. Optimization:

   • Disable secondary upscale for 2x speed boost

   • Use --medvram flag for low-VRAM GPUs