1. Workflow Overview

This workflow, named "万物光绘-光线-描边特效", is designed to add light painting and edge glow effects to input images, creating enhanced visuals with sci-fi aesthetics. Key features include:

• Depth map generation via Depth Anything V2

• Light painting effects powered by custom LoRAs

• Tiled processing for high-resolution output

• Final upscaling to HD quality

Core Models:

• Stable Diffusion: Base image generation

• Depth Anything V2: Depth estimation

• 4x_NMKD-Siax_200k: Upscaling model

• Custom LoRAs:

  • 阿容容_F.1 光绘描边 (weight=0.8)

  • 指鹿AI_蛇年_强质感 (weight=0.2)

2. Node Details

Key Nodes:

1. DepthAnything_V2: Generates depth maps from input images

   • Install: Via ComfyUI Manager (search "Depth Anything")

   • Model: depth_anything_v2_vitl_fp32.safetensors (auto-downloaded)

2. InstructPixToPixConditioning: Merges depth maps with text prompts

   • Input: Depth map + positive/negative prompts

   • Output: Conditioned latent data

3. LoraLoaderModelOnly: Loads light-painting LoRAs

   • Path: models/loras/

   • Manual setup required for LoRA files

4. TTP_Image_Tile_Batch & TTP_Image_Assy: Tile-based processing

   • Splits large images into patches for VRAM efficiency

   • Install: Requires Tiled Diffusion plugin (GitHub)

5. FluxGuidance: Dynamic guidance scaling (guidance_scale=30)

3. Workflow Structure

Group 1: Input Control

• Input: Source image (e.g., 北京央视大楼.jpg)

• Process:

  • Depth map generation

  • Resize to 1024x1024

• Output: Depth map + normalized image

Group 2: Text Prompts

• Input:

  • Positive prompt (e.g., "Blue-gold light trails, sci-fi lines, 8K")

  • Negative prompt (e.g., "blurry, low quality")

• Process: CLIP text encoding

Group 3: Sampling

• Model: Base F.1_Depth-fp16 + LoRAs

• Sampler: Euler, 30 steps, CFG=7.0

• Output: Latent image with light effects

Group 4: Upscale & Refine

• Input: Initial output

• Process:

  • 4x upscaling (4x_NMKD-Siax)

  • Tiled processing for stability

• Output: HD image (e.g., 1920x1920)

4. Inputs & Outputs

• Inputs:

  • Image: Any size (recommended ≥512x512)

  • Prompts: Describe light effects (e.g., "glow, dynamic lines")

  • Seed: Optional fixed value (e.g., 316430325547060)

• Output:

  • Path: ComfyUI/output/

  • Format: PNG (HD with light effects)

5. Notes

1. VRAM: ≥12GB GPU (tiling reduces usage)

2. Dependencies:

   • Verify LoRA and Depth Anything models are installed

3. Troubleshooting:

   • Missing models: Check models/loras/ and models/depth_anything/

   • OOM errors: Reduce tile size (e.g., ≤1024x1024)

4. Optimization:

   • Enable xformers

   • Lower FluxGuidance value if slow