1. Workflow Overview

This workflow converts realistic photos into American retro comic-style line art through:

• Depth extraction via DepthAnything V2

• Style transfer using Flux model with LoRAs

• Tiled processing + upscaling for high-quality output

Core Models:

• DepthAnything_V2: Extracts image depth structure

• Flux Dev Model: Base generator (with American Colorful Illustration_F.1 and Retro Comic_F.1 LoRAs)

• 4x-UltraSharp: Upscaling model

2. Key Nodes

Required Custom Nodes:

1. Impact Pack (Install via ComfyUI Manager):

   • Provides ImpactImageBatchToImageList etc.

2. DepthAnything (Manual install):

   • Download model depth_anything_v2_vitl_fp32.safetensors

3. FLUX Nodes:

   • FluxGuidance: Controls conditioning strength

   • Flux Dev Model: Requires separate download

Dependencies:

• Place LoRAs in models/loras/:

  • American Colorful Illustration_F.1 (weight 0.25)

  • Retro Comic_F.1 (weight 0.5)

3. Workflow Structure

Group Logic:

1. Input Preprocessing:

   • Nodes: LoadImage → ImageResizeKJ → DepthAnything_V2

   • Input: Any image (e.g., city-6809824_1920.jpg)

   • Output: 1024x1024 depth map

2. Conditioning:

   • Nodes: CLIPTextEncode + ConditioningMultiCombine

   • Key Prompt:

     American retro comic art, line contours, anime, no text, no characters  

3. Tiled Generation:

   • Nodes: TTP_Image_Tile_Batch → KSampler → VAEDecodeTiled

   • Parameters: Tile size 1024x1024, overlap 64px

4. Post-Processing:

   • Nodes: ImageUpscaleWithModel → LayerColor: Brightness & Contrast

   • Action: 4x upscale + contrast adjustment

4. Input & Output

Inputs:

• Required: Source image (via LoadImage)

• Optional:

  • Seed value (random if empty)

  • Style strength (adjust LoRA weights)

Output:

• Format: PNG

• Resolution: 7680x7680 (example)

5. Notes

1. Hardware:

   • Recommended ≥12GB VRAM (tiling still demands high memory)

2. Troubleshooting:

   • DepthAnything model missing: Manually download to models/depth_anything/

   • LoRA load failures: Check filename special characters

3. Optimization:

   • Reduce KSampler steps (default 20 → 15)

   • Use easy cleanGpuUsed to free VRAM