1. Workflow Overview

This workflow is designed for portrait photography, combining Stable Diffusion with a LoRA (人像光影粒子调节器_L1.06) to generate high-quality portraits featuring artistic lighting and particle effects. It supports both online (default) and local execution modes.

2. Core Models

• Main Model: 高级质感人像-FLUX_v1.0 (Download)

  • Function: Base image generation with enhanced portrait details.

• LoRA Model: 人像光影粒子调节器_L1.06 (Download)

  • Function: Adds dynamic light/shadow and particle effects (e.g., hair highlights, bokeh).

3. Key Nodes

1. CheckpointLoaderSimple

   • Role: Loads the main model (高级质感人像-FLUX_v1.0).

   • Installation: Built-in node.

2. LoraLoader

   • Role: Applies the LoRA with default strength 0.8.

3. CLIPTextEncode

   • Role: Encodes prompts (e.g., "Asian woman in forest with glowing particles").

4. KSampler

   • Role: Controls sampling (e.g., euler, steps=30, cfg=1).

5. VAEDecode

   • Role: Decodes latent images into final output.

6. RandomNoise

   • Role: Generates initial noise for randomness.

4. Workflow Structure (Groups)

• Online Group (Default):

  • Full pipeline: Model → Prompt → Sampling → Decoding.

  • Input: Text prompts, resolution (768x1024).

  • Output: Portrait with light/particle effects.

• Local Group (Manual Activation):

  • Similar structure but may include local optimizations.

5. Inputs & Outputs

• Inputs:

  • Positive prompt (e.g., "Sunlit Asian woman with ethereal particles").

  • Resolution: 768x1024, seed (random/fixed).

• Output:

  • High-resolution portrait (PNG/JPG) with artistic effects.

6. Notes

1. Model Dependencies:

   • Download models and place them in ComfyUI/models.

2. Performance:

   • GPU ≥8GB VRAM recommended. Use euler for speed/quality balance.

3. Troubleshooting:

   • If prompts fail, check CLIPTextEncode connections.

   • Reduce LoRA weight (e.g., 0.5) if effects are too strong.