1. Workflow Overview

This professional workflow transforms real photos into 3D figurines, featuring:

• IPAdapter-based feature preservation

• FaceDetailer enhancement

• Dual-style outputs (moe style + dark fantasy)

• Professional figurine parameter control

2. Node Breakdown

Key Nodes:

1. IPAdapterAdvanced

   • Strength: 0.88

   • Mode: concat

2. FaceDetailer

   • Multi-model collaboration:

     • SAM_vit_h + YOLOv8m-seg

   • Refinement: 20 steps

3. Dual Text Encoders

   • Moe style prompts

   • Dark fantasy prompts

Dependencies:

• ComfyUI-IPAdapter-Plus required

• ~8GB model downloads

3. Workflow Stages

1. Feature Extraction

   • CLIP vision encodes real photo

   • IPAdapter binds features

2. Generation

   • Empty latent: 1024x1024

   • dpmpp_2m_alt sampler

3. Post-Processing

   • 512px face refinement

   • Dual output channels

4. I/O Specification

Inputs:

• Source image: Half-body portrait

• Style selection via prompts

Outputs:

• Path: /ComfyUI/

• Resolution: 1024x1024

5. Critical Notes

1. Hardware

   • VRAM ≥12GB

   • Recommended: RTX 3090/4090

2. Model Paths

   • Figurine models: ComfyUI/models/checkpoints/

   • IPAdapter: ComfyUI/models/ipadapter/

3. Troubleshooting

   • Face distortion:

     Adjust mask_threshold (default:0.93)  

   • Weak style transfer:

     Increase IPAdapter strength