1. Workflow Overview

• Purpose: Transforms line art into colored illustrations or 3D-style renders using ControlNet-guided generation and super-resolution.

• Key Techniques:

  • ControlNet Canny: Strictly preserves line art edges.

  • Dual CLIP Encoding: Enhances complex prompt (e.g., materials, lighting) understanding.

  • UltimateSDUpscale: Tile-based upscaling for 4K output.

2. Core Models & Nodes

3. Workflow Structure

Group 1: Line Art Preprocessing

• Input: User-uploaded line art (e.g., .jpg).

• Output: Resized image + Canny edges.

• Key Nodes:

  • CannyEdgePreprocessor: Thresholds 100/200 (adjustable).

Group 2: ControlNet-Guided Generation

• Input: Canny edges + prompt (e.g., "Vibrant digital illustrations...").

• Output: Latent image constrained by line art.

• Key Nodes:

  • ControlNetApplyAdvanced: Weight=0.8.

Group 3: Image Generation

• Parameters: 768x1024, Euler sampler, denoise=1.0.

Group 4: Ultra HD Upscaling (Right-Click Enable)

• Tile Processing: 8 tiles at 1352x1352 each.

• Detail Refinement: Secondary sampling (denoise=0.28).

4. Inputs & Outputs

• Inputs:

  • Required: Line art image, positive prompt (auto-translated to English).

• Outputs:

  • Colored/3D image (PNG).

  • 4K upscaled version (manual trigger).

5. Notes

1. Line Art Quality: Clear, closed lines recommended.

2. VRAM: ≥12GB for 4K upscaling.

3. Troubleshooting:

   • Color bleeding? Increase ControlNet weight (0.9~1.0).

   • Blurry details? Raise denoise in UltimateSDUpscale (0.3~0.5).