1. Workflow Overview

This workflow generates high-quality child ID photos by combining face detection, style transfer, and LoRA fine-tuning. It supports multiple clothing templates and background options, ideal for school or passport photos.

2. Core Models

• Main Model: 基础算法_F.1

  • Function: Base image generation with portrait optimization.

• LoRA Model: 儿童证件照 | 人像摄影系列_v1.0

  • Function: Enhances child-specific features (skin tone, expressions).

• PulidFlux Model: pulid_flux_v0.9.1.safetensors

  • Function: Improves face similarity and detail retention.

3. Key Nodes

1. LoadImage

   • Role: Loads input images (e.g., casual photos) and auto-crops faces.

2. AutoCropFaces

   • Role: Crops faces to 1:1 ratio (default) for ID photo standards.

3. ApplyPulidFlux

   • Role: Adjusts face similarity weight (default 0.6).

4. LoraLoader

   • Role: Applies child portrait LoRA at strength 1.

5. CLIPTextEncode

   • Role: Inputs preset prompts (e.g., "Asian boy, white shirt, plain background").

6. VAEDecode

   • Role: Renders final image at 768x1024 resolution.

4. Workflow Structure (Groups)

• Preset Group:

  • Offers 4 prompt templates (e.g., school uniform, casual wear).

• Face Crop Group:

  • Uses AutoCropFaces with params like scale 1.5, padding 0.45.

• Generation Group:

  • Includes sampler (euler, 30 steps), VAE decode, and image save.

5. Inputs & Outputs

• Inputs:

  • Child photo (.jpg/.png), preferably frontal face.

  • Optional prompt templates (e.g., Template 1: School uniform).

• Output:

  • Standard ID photo (768x1024, white background) saved as .png.

6. Notes

1. Dependencies:

   • Download pulid_flux_v0.9.1.safetensors and LoRA to model folders.

2. Performance:

   • GPU ≥6GB VRAM recommended. Use euler for speed/quality balance.

3. Troubleshooting:

   • Low face similarity? Increase weight in ApplyPulidFlux (0.6~1.0).

   • Dirty background? Ensure prompts include "plain white background".