1. Overview

This workflow replicates RedNote viral images with AI, featuring:

• Style Transfer: Uses RedNote FP8 Model for authentic aesthetics.

• Auto-Prompting: Florence2 analyzes uploaded images to generate keywords.

• Face Swapping: PulidFlux replaces faces with user-uploaded portraits.

• High-Res Output: 1024x1536 resolution for platform compatibility.

Core Models:

• Main: 小红书自然真实FP8模型

• LoRA: Flux_小红书真实风格_V1 (weight=0.8)

• Prompt Model: Florence-2-large-PromptGen-v1.5

• Face Fusion: pulid_flux_v0.9.0.safetensors

2. Key Nodes

Dependencies:

• Plugins:

  • Florence2: GitHub.

  • PulidFlux: Download model manually.

• Models:

  • RedNote model: LibLib.art.

3. Workflow Groups

1. Input Phase:

   • XHS Image Upload: Reference image for style analysis.

   • User Portrait: For face replacement.

2. Processing:

   • Prompt Generation: Florence2Run → StringFunction adds prefixes (e.g., xhs,).

   • Face Fusion: PulidFlux processes user portrait.

3. Generation:

   • Sampling: 1024x1536, 25 steps, CFG=3.5.

4. Inputs & Outputs

• Inputs:

  • Reference Image: 1+ RedNote-style images.

  • User Portrait: High-quality front-facing photo.

• Output: HD image saved to ComfyUI/output.

5. Notes

1. VRAM: ≥10GB GPU recommended (FP8 models reduce usage).

2. Face Swap Tips:

   • Use well-lit portraits without obstructions.

3. Troubleshooting:

   • Florence2 download error: Manually download models.

   • PulidFlux dependency: Install insightface via pip.

4. Optimization: Lower resolution (e.g., 768x1152) if OOM occurs.