1. Workflow Overview

• Purpose:
  Generates customizable cartoon emoji packs with:

  • 4 expressions of a 2-year-old girl (e.g., joyful, speaking, shy).

  • Facial detail enhancement via PuLID-Flux.

  • Style transfer using Redux models.

• Core Models:

  • Stable Diffusion (基础算法_F.1): Base image generation.

  • PuLID-Flux (pulid_flux_v0.9.0.safetensors): Improves facial consistency.

  • Redux Style Model (flux1-redux-dev): Adds cartoonish effects.

2. Key Nodes

• Critical Nodes:

  • PulidFluxModelLoader: Loads PuLID-Flux (manual install required).

  • DetailDaemonSamplerNode: Dynamically adjusts sampling parameters.

  • ReduxAdvanced: Post-processing for stylization.

• Dependencies:

  • Models:

    • PuLID-Flux: Download from HuggingFace.

    • Redux: Place in models/style_model.

  • Plugins: Install DetailDaemon via ComfyUI Manager.

3. Workflow Structure

• Group 1: Text-to-Image

  • Input: Text prompts (e.g., "2-year-old girl, happy expression").

  • Nodes: CLIPTextEncode, UNETLoader.

  • Output: Latent image.

• Group 2: Facial Enhancement

  • Input: Reference image (loaded via LoadImage).

  • Nodes: ApplyPulidFlux.

  • Output: Optimized model output.

• Group 3: Post-Processing

  • Nodes: VAEDecode, ReduxAdvanced.

  • Output: Final emoji images (PNG).

4. Input & Output

• Global Inputs:

  • Text prompts (4 expressions).

  • Reference image.

  • Resolution: 1024x1024 (EmptyLatentImagePresets).

• Final Output: 4 emoji images (saved to ComfyUI/output).

5. Notes

• Errors:

  • Missing PuLID-Flux: Download manually.

  • InsightFace: Install via pip install insightface.

• Optimization:

  • Use CPU mode for InsightFace.

  • Reduce resolution for lower VRAM usage.

• Compatibility:

  • Requires ComfyUI ≥0.4 and NVIDIA GPU (≥8GB VRAM).