1. Workflow Overview

This advanced workflow specializes in precise style transfer using depth control and AI-powered image generation. It combines:

• Depth-guided image generation

• AI style transfer

• Automatic prompt generation

• High-quality output rendering

Key applications include portrait enhancement, artistic style transfer, and controlled image generation.

2. Core Models

3. Key Components

4. Workflow Structure

1. Input Section:

   • "照片——放这里": Source image input (658×1170)

   • "风格参考图——放这里": Style reference image (452×600)

2. Prompt Generation:

   • Uses Llama-3 to analyze images and generate descriptive prompts

3. Depth Control:

   • Creates depth maps for controlled generation

   • Processes at 1216px resolution

4. Style Transfer:

   • Applies style model with CLIP vision encoding

   • Uses flux1-redux-dev style model

5. Generation:

   • 8 sampling steps with Euler method

   • 1216px output resolution

5. Input/Output

Inputs:

• Source image (PNG/JPG)

• Style reference image (optional)

• Prompt (can be AI-generated or manual)

Outputs:

• Stylized image with depth control

• Intermediate depth maps

• Generated prompts

6. Technical Notes

• Requires significant VRAM (recommended 12GB+)

• Uses fp8 precision for some models

• Depth processing at 1216px may be memory-intensive

• Style model applies multiply blending

7. Installation Requirements

1. Required custom nodes:

   • ComfyUI-ControlNet-Aux (for depth processor)

   • Derfuu's Modded Nodes (for smart scaling)

   • Joy_caption_two (for prompt generation)

2. Model downloads needed:

   • depth_anything_v2_vitl.pth

   • flux1-redux-dev style model

   • HyperL-F.1 LoRA