1. Workflow Overview

This workflow automatically converts hand-drawn sketches into 3D-style renders with two output modes:

• Realistic 3D rendering (e.g., game character models)

• Clay-style gray model (low-poly aesthetic)
  Key features: ControlNet structure control, LoRA style tuning, auto-prompt generation, and image upscaling.

2. Core Models

3. Key Nodes

• ControlNetApplyAdvanced (Node 35/52)
  ▶ Purpose: Controls generation structure via depth/lineart
  ▶ Requires: comfyui_controlnet_aux extension

• LoRA Loaders (Node 15/28/73)
  ▶ Purpose: Dynamically switches 3D styles (adjustable strength)
  ▶ Tip: Total strength ≤1.8 when stacking multiple LoRAs

• Florence2DescribeImage (Node 40)
  ▶ Purpose: AI-generated prompts from input images
  ▶ Dependency: comfyui_caption_this plugin

4. Workflow Structure

Group 1: Input Preprocessing

• Input: Sketch (e.g., 2025-05-14 13-33-43-50.png)

• Output: Normalized 512x768 image

• Key Nodes: ImageResize+ (Node 36) + Zoe-DepthMapPreprocessor (Node 33)

Group 2: Dual-Path Generation

• Path A (3D Render):

  • Uses 3DMM_V12 LoRA + Depth ControlNet

  • Output: /3D/A_Render_

• Path B (Clay Style):

  • Uses GrayClay_V1.5.5 LoRA + Lineart ControlNet

  • Output: /3D/A_GrayModel_

Group 3: Post-Processing

• Adds stroke effects via LayerStyle nodes

• Final composite saved as /3D_Cover/A_

5. Input/Output

• Required Inputs:

  • Sketch image (PNG/JPG)

  • Optional prompts (auto-generated by default)

  • Seed value (default: 2048)

• Outputs:

  • Two 3D-style renders

  • Composite cover image with effects

6. Notes

⚠️ VRAM: 8GB+ recommended. Use --medvram for lower consumption.
⚠️ LoRA Conflicts: Manually adjust strengths when stacking (see Node 30 notes).
⚠️ Optimization: Disable upscale group during sampling (Node 110 tip).
💡 Recommended Settings:

• Sampler: Euler a

• Steps: 8-20

• CFG Scale: 2.0-2.8